<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Dr. Premkumar Borugadda, M.Tech.,Ph.D., Postdoc on Medium]]></title>
        <description><![CDATA[Stories by Dr. Premkumar Borugadda, M.Tech.,Ph.D., Postdoc on Medium]]></description>
        <link>https://medium.com/@premkumar.jones?source=rss-aa72d89a5ceb------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*bSQQ58JaY0W1hh5ehqFwtg.jpeg</url>
            <title>Stories by Dr. Premkumar Borugadda, M.Tech.,Ph.D., Postdoc on Medium</title>
            <link>https://medium.com/@premkumar.jones?source=rss-aa72d89a5ceb------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 22:03:09 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@premkumar.jones/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Best Machine Learning and Deep Learning Models for Agronomic Research]]></title>
            <link>https://medium.com/@premkumar.jones/best-machine-learning-and-deep-learning-models-for-agronomic-research-d6a2e3d93dc7?source=rss-aa72d89a5ceb------2</link>
            <guid isPermaLink="false">https://medium.com/p/d6a2e3d93dc7</guid>
            <dc:creator><![CDATA[Dr. Premkumar Borugadda, M.Tech.,Ph.D., Postdoc]]></dc:creator>
            <pubDate>Sun, 20 Sep 2026 07:54:32 GMT</pubDate>
            <atom:updated>2026-09-20T07:54:32.605Z</atom:updated>
            <content:encoded><![CDATA[<h3>Choosing the Right Model for Agricultural Data, Prediction and Scientific Interpretation</h3><p>The integration of Machine Learning (ML), Deep Learning (DL), and Artificial Intelligence (AI) is transforming modern agronomic research. These technologies enable researchers to analyze complex relationships among crops, soil, weather, water, nutrients, management practices, and environmental conditions.</p><p>Today, ML and DL models are being applied to:</p><ul><li>Crop-yield prediction</li><li>Soil-property mapping</li><li>Fertilizer recommendation and nutrient management</li><li>Crop disease and pest detection</li><li>Irrigation scheduling and water-stress assessment</li><li>Weed identification</li><li>Remote-sensing image analysis</li><li>Crop-growth monitoring</li><li>Drought and climate-risk assessment</li><li>Plant phenotyping</li><li>Agricultural management-zone delineation</li><li>Decision-support and precision-farming systems</li></ul><p>However, no single model is universally best for every agronomic problem. Model selection should be guided by the research question, dataset size, input type, experimental design, spatial and temporal structure, interpretability requirements, and intended application—not simply by choosing the most advanced algorithm.</p><h3>1. Regression Models</h3><h3>Linear and Multiple Linear Regression</h3><p>Regression models are valuable baselines for studying relationships between agronomic variables.</p><p><strong>Applications:</strong></p><ul><li>Estimating the relationship between fertilizer rate and crop yield</li><li>Examining the effects of rainfall and temperature on productivity</li><li>Predicting biomass from vegetation indices</li><li>Evaluating relationships among soil nutrients, crop growth and yield</li><li>Estimating crop water requirements</li></ul><p><strong>Example:</strong> Predicting maize yield using nitrogen application rate, rainfall, soil organic carbon and planting density.</p><p>Their primary advantage is interpretability. Researchers can directly examine how each independent variable is associated with the predicted outcome.</p><h3>Ridge, Lasso and Elastic Net Regression</h3><p>These regularized regression models are useful when datasets contain many correlated variables.</p><p><strong>Applications:</strong></p><ul><li>Selecting important soil and weather variables</li><li>Managing multicollinearity among vegetation indices</li><li>Identifying important spectral bands</li><li>Developing interpretable yield-prediction models</li><li>Selecting influential nutrient and management variables</li></ul><p>Lasso can reduce some coefficients to zero, making it useful for feature selection. Ridge controls unstable coefficients, while Elastic Net combines the advantages of both approaches.</p><h3>Generalized Linear Models</h3><p>Generalized Linear Models extend ordinary regression to outcomes that do not follow a normal distribution.</p><p><strong>Applications:</strong></p><ul><li>Binary disease occurrence</li><li>Pest-count analysis</li><li>Germination success</li><li>Crop-failure probability</li><li>Presence or absence of nutrient deficiency</li></ul><p>For example, logistic regression can estimate the probability of disease occurrence using temperature, humidity and rainfall.</p><h3>2. Random Forest</h3><p>Random Forest is one of the most widely used algorithms in agricultural and environmental research. It combines predictions from many decision trees to produce a stable result.</p><p><strong>Applications:</strong></p><ul><li>Crop-yield prediction</li><li>Soil organic carbon estimation</li><li>Soil-moisture prediction</li><li>Nutrient-status classification</li><li>Land-use and crop-type mapping</li><li>Disease-risk prediction</li><li>Feature-importance analysis</li></ul><p><strong>Example:</strong> Predicting rice yield using rainfall, temperature, soil pH, nitrogen content, irrigation frequency and satellite-derived NDVI.</p><p>Random Forest performs well with nonlinear relationships, mixed variables and complex interactions. It is also relatively resistant to overfitting when properly configured.</p><h3>3. Extra Trees</h3><p>Extremely Randomized Trees, commonly called Extra Trees, introduce more randomness while constructing decision trees.</p><p><strong>Applications:</strong></p><ul><li>Crop-yield and biomass prediction</li><li>Soil-property estimation</li><li>Feature ranking</li><li>Classification of crops and land-cover types</li><li>Analysis of high-dimensional tabular datasets</li></ul><p>Extra Trees can sometimes provide faster training and better generalization than Random Forest, although performance depends on the dataset.</p><h3>4. XGBoost, LightGBM and CatBoost</h3><p>Gradient-boosting algorithms build decision trees sequentially, with each new tree attempting to correct previous errors.</p><h3>XGBoost</h3><p>XGBoost is powerful for structured and tabular agronomic datasets.</p><p><strong>Applications:</strong></p><ul><li>Yield prediction</li><li>Fertilizer-response modelling</li><li>Disease-risk assessment</li><li>Soil-fertility prediction</li><li>Crop suitability analysis</li><li>Weather-related risk prediction</li></ul><h3>LightGBM</h3><p>LightGBM is efficient for large datasets and can train faster than many conventional boosting methods.</p><p><strong>Applications:</strong></p><ul><li>Large-scale soil mapping</li><li>Regional crop-yield forecasting</li><li>Agricultural monitoring using multiple satellite and climate variables</li><li>High-dimensional farm-management datasets</li></ul><h3>CatBoost</h3><p>CatBoost is especially useful when datasets contain several categorical variables.</p><p><strong>Applications:</strong></p><ul><li>Farm-level datasets containing crop variety, soil type and management category</li><li>Yield prediction using numerical and categorical inputs</li><li>Agricultural survey data</li><li>Fertilizer and cultivar recommendation</li></ul><p><strong>Example:</strong> Predicting wheat yield using soil properties, rainfall, temperature, cultivar, irrigation method and farm-management practice.</p><h3>5. Support Vector Machine and Support Vector Regression</h3><p>Support Vector Machine (SVM) is used for classification, while Support Vector Regression (SVR) is used for continuous prediction.</p><p><strong>Applications:</strong></p><ul><li>Crop and soil classification</li><li>Disease detection</li><li>Yield prediction</li><li>Quality grading</li><li>Drought-stress classification</li><li>Spectral-data analysis</li></ul><p>SVM and SVR can perform well on small-to-medium datasets with complex nonlinear relationships. They are particularly valuable when the number of features is large compared with the number of observations.</p><p>However, their performance is sensitive to feature scaling, kernel selection and hyperparameter tuning.</p><h3>6. K-Nearest Neighbours</h3><p>K-Nearest Neighbours predicts an observation based on the most similar observations in the training dataset.</p><p><strong>Applications:</strong></p><ul><li>Soil-type classification</li><li>Crop classification</li><li>Disease identification</li><li>Seed-quality classification</li><li>Similar-field identification</li></ul><p>KNN is easy to understand and useful as a baseline, but it can become computationally expensive with large datasets and may perform poorly when irrelevant variables are included.</p><h3>7. Artificial Neural Networks and Multilayer Perceptrons</h3><p>Artificial Neural Networks (ANNs), including Multilayer Perceptrons (MLPs), can model complex nonlinear relationships.</p><p><strong>Applications:</strong></p><ul><li>Crop-yield prediction</li><li>Biomass estimation</li><li>Soil-moisture prediction</li><li>Nutrient prediction</li><li>Evapotranspiration estimation</li><li>Irrigation-requirement forecasting</li><li>Greenhouse environmental control</li></ul><p><strong>Example:</strong> Estimating crop water requirements using temperature, humidity, solar radiation, wind speed, soil moisture and crop-growth stage.</p><p>ANNs are powerful but generally require more data and tuning than conventional ML models. Their internal reasoning can also be difficult to interpret without explainability methods.</p><h3>8. Convolutional Neural Networks</h3><p>Convolutional Neural Networks (CNNs) are designed to learn spatial features from images.</p><p><strong>Applications:</strong></p><ul><li>Leaf-disease classification</li><li>Weed identification</li><li>Fruit counting</li><li>Crop-quality inspection</li><li>Seed classification</li><li>Plant phenotyping</li><li>Pest detection</li><li>UAV and satellite-image classification</li></ul><p><strong>Example:</strong> Classifying healthy and diseased tomato leaves using EfficientNet, ResNet, DenseNet or MobileNet.</p><p>Popular CNN architectures include:</p><ul><li>LeNet</li><li>AlexNet</li><li>VGG16 and VGG19</li><li>ResNet</li><li>DenseNet</li><li>EfficientNet</li><li>MobileNet</li><li>Inception</li><li>ConvNeXt</li></ul><p>Transfer learning is especially valuable when the agricultural image dataset is relatively small.</p><h3>9. Object-Detection Models</h3><p>Image classification identifies the overall class of an image, whereas object-detection models locate and classify individual objects within it.</p><p>Popular models include:</p><ul><li>YOLO</li><li>Faster R-CNN</li><li>RetinaNet</li><li>SSD</li><li>DETR</li></ul><p><strong>Applications:</strong></p><ul><li>Fruit detection and counting</li><li>Weed localization</li><li>Pest identification</li><li>Detection of diseased leaf regions</li><li>Crop-row detection</li><li>Livestock monitoring</li></ul><p><strong>Example:</strong> Using YOLO to detect and count mangoes in orchard images captured by a drone.</p><h3>10. Image-Segmentation Models</h3><p>Segmentation models classify individual pixels, making them suitable for measuring the exact area and shape of agricultural objects.</p><p>Popular architectures include:</p><ul><li>U-Net</li><li>DeepLabV3+</li><li>Mask R-CNN</li><li>SegNet</li><li>SegFormer</li></ul><p><strong>Applications:</strong></p><ul><li>Disease-lesion segmentation</li><li>Weed-area mapping</li><li>Crop-canopy extraction</li><li>Root segmentation</li><li>Field-boundary detection</li><li>Flooded-area mapping</li><li>Soil and land-cover mapping</li></ul><p><strong>Example:</strong> Using U-Net to segment disease-affected regions in crop leaves and calculate disease severity.</p><h3>11. LSTM and GRU Models</h3><p>Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) networks are designed for sequential and time-series data.</p><p><strong>Applications:</strong></p><ul><li>Crop-growth forecasting</li><li>Yield prediction across seasons</li><li>Weather forecasting</li><li>NDVI time-series analysis</li><li>Soil-moisture forecasting</li><li>Irrigation-demand prediction</li><li>Pest and disease outbreak forecasting</li></ul><p><strong>Example:</strong> Predicting weekly NDVI using historical NDVI, rainfall, temperature and soil-moisture observations.</p><p>LSTM and GRU models can capture temporal dependencies but require sufficient sequential data and careful treatment of missing observations.</p><h3>12. Temporal Convolutional Networks</h3><p>Temporal Convolutional Networks use convolutional operations to analyze sequential data.</p><p><strong>Applications:</strong></p><ul><li>Weather and soil-moisture forecasting</li><li>Crop-growth monitoring</li><li>Irrigation-demand prediction</li><li>Multisensor agricultural time-series analysis</li></ul><p>They can sometimes train faster than recurrent models and capture long-term temporal patterns effectively.</p><h3>13. CNN–LSTM Models</h3><p>CNN–LSTM models combine spatial-feature extraction with temporal learning.</p><p><strong>Applications:</strong></p><ul><li>Monitoring crop development using multitemporal satellite imagery</li><li>In-season yield forecasting</li><li>Disease progression analysis</li><li>Crop-growth-stage identification</li><li>Spatiotemporal drought assessment</li></ul><p><strong>Example:</strong> A CNN extracts spatial features from a sequence of Sentinel-2 images, while an LSTM learns how those features change during the growing season.</p><h3>14. Transformer Models</h3><p>Transformers use attention mechanisms to identify important relationships across long sequences or large datasets.</p><p><strong>Applications:</strong></p><ul><li>Long-term agricultural time-series forecasting</li><li>Large-scale remote-sensing analysis</li><li>Crop-type mapping</li><li>Multimodal data integration</li><li>Climate–crop interaction modelling</li><li>Agricultural text and advisory systems</li></ul><p>Examples include:</p><ul><li>Vision Transformer</li><li>Swin Transformer</li><li>Temporal Fusion Transformer</li><li>Time-Series Transformer</li><li>SegFormer</li><li>DETR</li></ul><p>Transformers can combine satellite imagery, weather data, soil information and field-management records. However, they generally require substantial data, computing resources and careful validation.</p><h3>15. Graph Neural Networks</h3><p>Graph Neural Networks model data represented as connected nodes and relationships.</p><p><strong>Applications:</strong></p><ul><li>Modelling water flow across connected fields</li><li>Irrigation-network analysis</li><li>Spatial modelling of farms and neighbouring plots</li><li>Disease-spread analysis</li><li>Agricultural supply-chain modelling</li><li>Soil-sampling networks</li></ul><p><strong>Example:</strong> Representing agricultural fields as nodes and spatial adjacency as edges to predict the regional spread of crop disease.</p><h3>16. Gaussian Process Regression</h3><p>Gaussian Process Regression provides predictions together with uncertainty estimates.</p><p><strong>Applications:</strong></p><ul><li>Soil-property mapping</li><li>Spatial interpolation</li><li>Crop-yield estimation</li><li>Environmental monitoring</li><li>Decision-making under uncertainty</li></ul><p>This method is particularly useful when observations are limited and prediction uncertainty is scientifically important. Its computational cost can become high for very large datasets.</p><h3>17. Bayesian Models</h3><p>Bayesian models combine previous knowledge with observed evidence and produce probability distributions rather than only point predictions.</p><p><strong>Applications:</strong></p><ul><li>Yield prediction under uncertainty</li><li>Fertilizer-response modelling</li><li>Climate-risk assessment</li><li>Pest-outbreak probability</li><li>Agricultural decision-making</li><li>Combining expert knowledge with experimental data</li></ul><p>Bayesian hierarchical models are especially useful when data come from multiple farms, seasons, locations or experimental blocks.</p><h3>18. Mixed-Effects Models</h3><p>Agronomic data often have hierarchical or repeated structures — for example, measurements from several plots, farms, years and locations.</p><p>Mixed-effects models can represent both fixed and random effects.</p><p><strong>Applications:</strong></p><ul><li>Multi-location field trials</li><li>Repeated crop measurements</li><li>Genotype × environment studies</li><li>Fertilizer experiments across seasons</li><li>Farm-level variability analysis</li></ul><p>These models should be considered when the experimental design contains blocks, replications, locations or repeated observations.</p><h3>19. K-Means and Other Clustering Methods</h3><p>Clustering is an unsupervised learning approach that discovers natural groups in data without predefined labels.</p><p>Common methods include:</p><ul><li>K-means</li><li>Hierarchical clustering</li><li>DBSCAN</li><li>Gaussian mixture models</li><li>Fuzzy C-means</li></ul><p><strong>Applications:</strong></p><ul><li>Creating agricultural management zones</li><li>Grouping soils according to fertility characteristics</li><li>Classifying farms by production patterns</li><li>Detecting spatial variability</li><li>Grouping crop genotypes</li><li>Identifying abnormal observations</li></ul><p><strong>Example:</strong> Dividing a field into low-, medium- and high-productivity management zones using soil electrical conductivity, elevation, NDVI and historical yield.</p><h3>20. Principal Component Analysis</h3><p>Principal Component Analysis (PCA) reduces many correlated variables to a smaller number of components.</p><p><strong>Applications:</strong></p><ul><li>Reducing correlated soil variables</li><li>Compressing vegetation indices</li><li>Exploring hyperspectral data</li><li>Visualizing genotype differences</li><li>Reducing model complexity</li><li>Controlling multicollinearity</li></ul><p>PCA is useful for exploration and dimensionality reduction, but the resulting components may not always have a simple agronomic interpretation.</p><h3>21. Partial Least Squares Regression</h3><p>Partial Least Squares Regression is valuable when predictors are numerous and highly correlated.</p><p><strong>Applications:</strong></p><ul><li>Hyperspectral crop analysis</li><li>Soil spectroscopy</li><li>Nutrient estimation</li><li>Biomass prediction</li><li>Quality-trait analysis</li></ul><p>It is commonly used in spectroscopy because it reduces predictor dimensionality while considering the response variable.</p><h3>22. Autoencoders</h3><p>Autoencoders are neural networks that learn compressed representations of complex data.</p><p><strong>Applications:</strong></p><ul><li>Reducing hyperspectral-image dimensions</li><li>Detecting anomalous crop conditions</li><li>Removing noise from sensor data</li><li>Learning features from unlabeled agricultural imagery</li><li>Compressing high-dimensional remote-sensing data</li></ul><p>Variational autoencoders can also generate synthetic representations and model data variability.</p><h3>23. Ensemble and Hybrid Models</h3><p>Ensemble models combine predictions from multiple algorithms.</p><p><strong>Examples:</strong></p><ul><li>Random Forest + XGBoost</li><li>CNN + SVM</li><li>CNN + LSTM</li><li>Remote-sensing model + weather model</li><li>Weighted averaging of RF, SVR and ANN</li><li>Stacking multiple base learners with a meta-model</li></ul><p>Ensembles may improve accuracy and stability, but additional complexity should be justified through consistent performance improvements and scientific relevance.</p><h3>24. Explainable AI for Agronomic Research</h3><p>High predictive accuracy alone is not sufficient for scientific research. Researchers must understand why a model produces a particular prediction.</p><p>Important explainability methods include:</p><ul><li>SHAP</li><li>Permutation feature importance</li><li>Partial dependence plots</li><li>Accumulated local effects</li><li>LIME</li><li>Saliency maps</li><li>Grad-CAM</li><li>Attention visualization</li></ul><p><strong>Examples:</strong></p><ul><li>SHAP can show how rainfall, nitrogen, soil moisture and NDVI contribute to predicted yield.</li><li>Grad-CAM can highlight the leaf regions used by a CNN to identify a disease.</li><li>Partial dependence plots can reveal how predicted yield changes across fertilizer levels.</li></ul><p>Explainability helps connect model predictions with agronomic processes, expert knowledge and practical management decisions.</p><h3>How to Select an Appropriate Model</h3><p>Research problem: Suitable starting modelsSmall experimental dataset Linear regression, regularized regression, SVM or Random Forest Structured farm or soil data Random Forest, XGBoost, LightGBM or CatBoostData containing many categoriesCatBoost or carefully encoded tree-based modelsMulti-location or repeated field trialsMixed-effects or Bayesian hierarchical modelsCrop-disease image classificationTransfer-learning CNNs or Vision TransformersWeed, fruit or pest localizationYOLO, Faster R-CNN or DETRDisease-area or canopy segmentationU-Net, DeepLabV3+ or SegFormerAgricultural time-series forecastingLSTM, GRU, TCN or Time-Series TransformerMultitemporal satellite imageryCNN–LSTM, 3D CNN or TransformerSoil spectroscopy or hyperspectral dataPLSR, PCA with ML, 1D CNN or TransformerAgricultural management zonesK-means, fuzzy clustering or DBSCANSpatial data with uncertaintyGaussian processes or Bayesian spatial modelsConnected fields or irrigation networksGraph Neural NetworksModel interpretationSHAP, permutation importance, PDP or ALE</p><h3>A Robust ML Workflow for Agronomic Research</h3><p>A scientifically sound workflow may be represented as:</p><p><strong>Research question → Experimental design and data collection → Data-quality assessment → Preprocessing → Exploratory analysis → Baseline statistical model → Feature engineering → ML/DL model development → Appropriate cross-validation → Performance evaluation → Uncertainty analysis → Explainability → Agronomic interpretation → External validation → Deployment or recommendation</strong></p><p>Important steps include:</p><ol><li>Clearly define the agronomic research question.</li><li>Verify sampling quality and experimental design.</li><li>Handle missing values, measurement errors and outliers carefully.</li><li>Avoid information leakage between training and testing datasets.</li><li>Establish a simple and interpretable baseline.</li><li>Compare multiple suitable models using identical data partitions.</li><li>Tune hyperparameters using only the training data.</li><li>Select validation methods that reflect the intended real-world application.</li><li>Report variability, uncertainty and confidence intervals where possible.</li><li>Interpret the findings using established agronomic knowledge.</li></ol><h3>Choosing the Correct Validation Strategy</h3><p>Random cross-validation is not appropriate for every agricultural dataset.</p><p>Researchers may need:</p><ul><li><strong>K-fold cross-validation</strong> for independent tabular observations</li><li><strong>Stratified cross-validation</strong> for imbalanced classification problems</li><li><strong>Group cross-validation</strong> when observations belong to the same farm, plot or subject</li><li><strong>Spatial cross-validation</strong> for soil and remote-sensing studies</li><li><strong>Temporal validation</strong> for forecasting future seasons</li><li><strong>Leave-one-location-out validation</strong> for multi-location field studies</li><li><strong>Leave-one-year-out validation</strong> for evaluating seasonal generalization</li><li><strong>External validation</strong> using a completely independent dataset</li></ul><p>If nearby pixels, repeated plot measurements or observations from the same farm appear in both training and testing sets, model performance may be seriously overestimated.</p><h3>Performance Metrics</h3><h3>Regression</h3><p>Common regression metrics include:</p><ul><li>Coefficient of determination (R²)</li><li>Root Mean Square Error (RMSE)</li><li>Mean Absolute Error (MAE)</li><li>Mean Absolute Percentage Error (MAPE)</li></ul><p>MAPE should be used carefully when actual values are zero or close to zero.</p><h3>Classification</h3><p>Common classification metrics include:</p><ul><li>Accuracy</li><li>Precision</li><li>Recall or sensitivity</li><li>Specificity</li><li>F1-score</li><li>ROC–AUC</li><li>Precision–Recall AUC</li><li>Cohen’s kappa</li><li>Matthews Correlation Coefficient</li></ul><p>Accuracy alone can be misleading when the classes are imbalanced.</p><h3>Segmentation and Object Detection</h3><p>Useful metrics include:</p><ul><li>Intersection over Union</li><li>Dice coefficient</li><li>Mean Average Precision</li><li>Precision and recall</li><li>Boundary-based accuracy</li><li>Counting error</li></ul><h3>Final Perspective</h3><p>A highly accurate model is not necessarily the most scientifically useful model.</p><p>In agronomic research, model selection should consider:</p><ul><li>Biological relevance</li><li>Agronomic interpretability</li><li>Experimental design</li><li>Data quality</li><li>Sample size</li><li>Spatial and temporal dependence</li><li>Prediction uncertainty</li><li>Computational requirements</li><li>Reproducibility</li><li>Practical usability</li></ul><p>A complex deep-learning model may achieve slightly higher accuracy, but a simpler model may provide clearer scientific insights, require fewer observations, and be easier to apply in real farming conditions.</p><p>The strongest agronomic studies do not merely report the model with the highest performance. They explain why the model works, where it may fail, whether it generalizes across locations and seasons, and how its results can support agricultural decision-making.</p><p><strong>Machine learning should not replace agronomic knowledge—it should strengthen our ability to understand, predict, and sustainably manage agricultural systems.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MO1ikBnlEUNtcBArkSWcGQ.png" /><figcaption>ML modes for Agronomic Research</figcaption></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d6a2e3d93dc7" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Computer Vision vs NLP vs Time Series vs Numerical Machine Learning]]></title>
            <link>https://medium.com/@premkumar.jones/computer-vision-vs-nlp-vs-time-series-vs-numerical-machine-learning-c4af5f4f1fab?source=rss-aa72d89a5ceb------2</link>
            <guid isPermaLink="false">https://medium.com/p/c4af5f4f1fab</guid>
            <category><![CDATA[computer-vision]]></category>
            <category><![CDATA[timeseries]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Dr. Premkumar Borugadda, M.Tech.,Ph.D., Postdoc]]></dc:creator>
            <pubDate>Tue, 08 Sep 2026 07:12:00 GMT</pubDate>
            <atom:updated>2026-09-08T07:12:00.779Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*axxd_kc6WnFzM1HPzvgsIA.png" /></figure><p>Machine learning is not a single method that works identically for every problem. The most appropriate approach depends largely on the structure of the available data.</p><p>An MRI scan, a clinical report, an ECG signal, and a table of laboratory measurements may all describe the same patient, but they represent information in fundamentally different forms. Therefore, each requires suitable preprocessing methods, feature representations, models, and evaluation metrics.</p><p>Four major machine-learning domains are:</p><ol><li>Computer Vision</li><li>Natural Language Processing</li><li>Time-Series Learning</li><li>Numerical or Tabular Machine Learning</li></ol><p>Understanding their differences is an important first step in selecting the right approach for a real-world problem.</p><p><strong>1. Computer Vision: Learning from Images and Videos</strong></p><p>Computer Vision enables machines to extract meaningful information from images and videos.</p><p>In a digital image, information is represented as a grid of pixel values. A grayscale image normally contains one intensity value for each pixel, whereas a colour image commonly contains red, green, and blue channel values.</p><p><strong>Common input data</strong></p><p>Computer Vision can process:</p><ul><li>Photographs</li><li>MRI and CT scans</li><li>X-ray images</li><li>Satellite and drone imagery</li><li>Microscopy images</li><li>Video frames</li><li>Thermal images</li><li>Agricultural field images</li></ul><p><strong>Common tasks</strong></p><p><strong>Image classification:</strong> Assigns one label to an entire image.</p><p>Example: Classifying a brain MRI as glioma, meningioma, pituitary tumour, or no tumour.</p><p><strong>Object detection:</strong> Identifies objects and their locations, usually using bounding boxes.</p><p>Example: Detecting vehicles in a traffic-monitoring image.</p><p><strong>Image segmentation:</strong> Assigns a class label to individual pixels.</p><p>Example: Separating a tumour region from healthy brain tissue.</p><p><strong>Image generation and enhancement:</strong> Creates, reconstructs, or improves visual content.</p><p>Example: Removing noise from a low-quality medical image.</p><p><strong>Common models</strong></p><p>Frequently used Computer Vision models include:</p><ul><li>Convolutional Neural Networks</li><li>ResNet</li><li>EfficientNet</li><li>DenseNet</li><li>Vision Transformer</li><li>Swin Transformer</li><li>YOLO</li><li>U-Net</li></ul><p><strong>Typical workflow</strong></p><p><strong>Image or video → Preprocessing → Visual features → CNN or Vision Transformer → Classification, detection, or segmentation</strong></p><p>Computer Vision therefore answers the simple question:</p><p><strong>What can the machine see in the image or video?</strong></p><p><strong>2. Natural Language Processing: Learning from Human Language</strong></p><p>Natural Language Processing, commonly called NLP, enables computers to analyse, understand, and generate human language.</p><p>Text cannot usually be provided directly to a machine-learning model. It must first be divided into tokens and converted into numerical representations known as vectors or embeddings.</p><p><strong>Common input data</strong></p><p>NLP systems work with:</p><ul><li>Documents</li><li>Sentences</li><li>Research articles</li><li>Emails</li><li>Reviews</li><li>Social-media posts</li><li>Medical reports</li><li>Clinical notes</li><li>Transcripts</li><li>Conversations</li><li>Speech converted into text</li></ul><p><strong>Common tasks</strong></p><p><strong>Text classification:</strong> Assigns a category to a document or sentence.</p><p>Example: Classifying an email as spam or legitimate.</p><p><strong>Sentiment analysis:</strong> Identifies whether an opinion is positive, negative, or neutral.</p><p><strong>Information extraction:</strong> Extracts names, dates, locations, diseases, measurements, or other entities from text.</p><p><strong>Machine translation:</strong> Converts text from one language into another.</p><p><strong>Text summarization:</strong> Produces a shorter version of a long document while preserving its main ideas.</p><p><strong>Question answering and generation:</strong> Produces responses or new text based on a prompt or context.</p><p><strong>Common models</strong></p><p>NLP approaches include:</p><ul><li>Recurrent Neural Networks</li><li>Long Short-Term Memory networks</li><li>Gated Recurrent Units</li><li>Transformers</li><li>BERT-family encoder models</li><li>T5 and other encoder–decoder models</li><li>Decoder-based large language models</li></ul><p><strong>Typical workflow</strong></p><p><strong>Text → Tokenization → Embeddings → Transformer or another NLP model → Classification, extraction, translation, summarization, or generation</strong></p><p>NLP primarily answers:</p><p><strong>What does the language mean, and how should the system respond?</strong></p><p><strong>3. Time-Series Learning: Learning from Changes Over Time</strong></p><p>Time-series learning focuses on observations recorded in chronological order.</p><p>Unlike an ordinary table in which rows may be independent, the order of observations is important in time-series data. A value recorded today may depend on values recorded yesterday, last week, or during the same season in previous years.</p><p><strong>Common input data</strong></p><p>Examples include:</p><ul><li>ECG and EEG signals</li><li>Weather measurements</li><li>Temperature and rainfall records</li><li>Stock prices</li><li>Energy demand</li><li>Industrial sensor readings</li><li>Internet of Things signals</li><li>Traffic volume</li><li>Crop-growth measurements</li><li>Satellite-derived vegetation-index sequences</li></ul><p><strong>Important characteristics</strong></p><p>Time-series data may contain:</p><ul><li>Trends</li><li>Seasonality</li><li>Cycles</li><li>Temporal dependence</li><li>Sudden changes</li><li>Missing timestamps</li><li>Noise</li><li>Irregular sampling intervals</li></ul><p>These properties must be examined before selecting a forecasting or classification model.</p><p><strong>Common tasks</strong></p><p><strong>Forecasting:</strong> Predicts future values from historical observations.</p><p>Example: Predicting next week’s electricity demand.</p><p><strong>Anomaly detection:</strong> Identifies unusual temporal behaviour.</p><p>Example: Detecting an abnormal pattern in an ECG signal.</p><p><strong>Temporal classification:</strong> Assigns a class to an entire sequence or a portion of it.</p><p>Example: Classifying human activities from wearable-sensor signals.</p><p><strong>Change-point detection:</strong> Identifies when the statistical behaviour of a sequence changes.</p><p><strong>Common models</strong></p><p>Time-series methods include:</p><ul><li>ARIMA and SARIMA</li><li>Exponential smoothing</li><li>Prophet</li><li>LSTM</li><li>GRU</li><li>Temporal Convolutional Networks</li><li>Time-series Transformers</li></ul><p>Traditional statistical models can perform very well on small, well-structured datasets. Deep-learning approaches become particularly useful for large, nonlinear, multivariate sequences.</p><p><strong>Typical workflow</strong></p><p><strong>Chronological observations → Temporal preprocessing → Temporal features → Statistical, recurrent, convolutional, or Transformer model → Forecasting, classification, or anomaly detection</strong></p><p>Time-series learning answers:</p><p><strong>What is happening over time, and what may happen next?</strong></p><p><strong>4. Numerical or Tabular Machine Learning: Learning from Structured Features</strong></p><p>Numerical or tabular machine learning works with structured datasets organised into rows and columns.</p><p>A row normally represents one observation, such as a patient, customer, crop field, transaction, or machine. Columns represent the characteristics of that observation.</p><p>Tabular data may contain both numerical and categorical variables.</p><p><strong>Common input features</strong></p><p>Examples include:</p><ul><li>Age</li><li>Income</li><li>Height and weight</li><li>Laboratory measurements</li><li>Soil properties</li><li>Temperature and pressure</li><li>Product dimensions</li><li>Transaction attributes</li><li>Crop type</li><li>Location category</li><li>Education level</li></ul><p><strong>Common tasks</strong></p><p><strong>Classification:</strong> Predicts a discrete category.</p><p>Example: Predicting whether a loan applicant is low-risk or high-risk.</p><p><strong>Regression:</strong> Predicts a continuous numerical value.</p><p>Example: Estimating soil organic carbon concentration.</p><p><strong>Clustering:</strong> Groups similar observations without predefined labels.</p><p>Example: Dividing customers into groups based on purchasing behaviour.</p><p><strong>Risk scoring and prediction:</strong> Estimates the probability of an event.</p><p>Example: Predicting the likelihood that a patient will develop a disease.</p><p><strong>Common models</strong></p><p>Popular tabular machine-learning algorithms include:</p><ul><li>Linear Regression</li><li>Logistic Regression</li><li>Support Vector Machine</li><li>Decision Tree</li><li>Random Forest</li><li>XGBoost</li><li>LightGBM</li><li>CatBoost</li><li>K-Nearest Neighbours</li><li>Multilayer Perceptron</li></ul><p>Tree-based ensemble models such as Random Forest, XGBoost, LightGBM, and CatBoost are often strong choices for structured datasets. However, model selection should always be supported by proper validation.</p><p><strong>Typical workflow</strong></p><p><strong>Rows and columns → Data cleaning and preprocessing → Feature engineering → Machine-learning algorithm → Classification, regression, clustering, or risk prediction</strong></p><p>Numerical machine learning answers:</p><p><strong>What relationship exists between the structured features and the target?</strong></p><p><strong>Key Differences at a Glance</strong></p><p><strong>Domain</strong></p><p><strong>Primary data</strong></p><p><strong>Common models</strong></p><p><strong>Major tasks</strong></p><p>Computer Vision</p><p>Images and videos</p><p>CNN, ResNet, EfficientNet, ViT, YOLO, U-Net</p><p>Classification, detection and segmentation</p><p>NLP</p><p>Text and human language</p><p>RNN, LSTM, Transformer, BERT and language models</p><p>Classification, extraction, translation, summarization and generation</p><p>Time Series</p><p>Chronologically ordered observations</p><p>ARIMA, LSTM, GRU, TCN and Transformers</p><p>Forecasting, temporal classification and anomaly detection</p><p>Numerical ML</p><p>Structured rows and columns</p><p>Regression, SVM, Random Forest, XGBoost and MLP</p><p>Classification, regression and clustering</p><p><strong>How Should We Select a Model?</strong></p><p>The best model should not be selected simply because it is new or popular. Model selection depends on several factors:</p><ul><li>Structure and quality of the data</li><li>Nature of the required output</li><li>Dataset size</li><li>Availability of labelled examples</li><li>Computational resources</li><li>Interpretability requirements</li><li>Real-time inference requirements</li><li>Cost of different types of errors</li><li>Selected evaluation metric</li><li>Deployment environment</li></ul><p>For example, a Transformer may be unnecessarily complex for a small tabular dataset. Similarly, an ordinary Random Forest cannot directly process raw images without an appropriate feature-extraction stage.</p><p>The correct question is not:</p><p>“Which algorithm is the most powerful?”</p><p>A better question is:</p><p>“Which representation and model are most suitable for this data, task, and operational requirement?”</p><p><strong>Can These Data Types Be Combined?</strong></p><p>Yes. Many real-world systems use more than one kind of data. This is generally called <strong>multimodal learning</strong> or <strong>feature fusion</strong>.</p><p>Consider a crop-monitoring system that uses:</p><ul><li>Satellite images through Computer Vision</li><li>Weather history through time-series learning</li><li>Soil measurements through tabular machine learning</li><li>Farmers’ written observations through NLP</li></ul><p>Each branch can extract a suitable representation from its own data. These representations are then combined to produce a final prediction or decision.</p><p><strong>Multimodal workflow</strong></p><p><strong>Multiple data types → Domain-specific preprocessing → Feature representations → Feature fusion → Prediction or decision</strong></p><p>Fusion can occur at different stages:</p><ul><li><strong>Early fusion:</strong> Combines raw or preprocessed features before modelling.</li><li><strong>Intermediate fusion:</strong> Combines learned representations inside a multimodal architecture.</li><li><strong>Late fusion:</strong> Combines the predictions produced by separate models.</li></ul><p>The appropriate fusion method depends on data availability, synchronization, model complexity, and the relationship between the modalities.</p><p><strong>An Easy Way to Remember the Four Domains</strong></p><ul><li><strong>Computer Vision → What do I see?</strong></li><li><strong>NLP → What does the language mean?</strong></li><li><strong>Time Series → What is changing over time?</strong></li><li><strong>Numerical ML → What relationships exist among structured features?</strong></li></ul><p>Another compact memory cue is:</p><p><strong>See → Understand → Track → Relate</strong></p><ul><li>See with Computer Vision.</li><li>Understand language with NLP.</li><li>Track changes using time-series learning.</li><li>Relate structured variables using numerical machine learning.</li></ul><p><strong>Conclusion</strong></p><p>Computer Vision, NLP, time-series learning, and numerical machine learning are not competing alternatives for every problem. They are specialised approaches designed for different data structures and objectives.</p><p>Computer Vision learns from spatial and visual patterns. NLP learns from language and context. Time-series methods learn from temporal behaviour. Numerical machine-learning models learn relationships within structured features.</p><p>In practical AI systems, these domains can also work together. The foundation of an effective solution is therefore not choosing the most fashionable algorithm — it is correctly understanding the data, defining the task, selecting meaningful evaluation measures, and building a model appropriate for the real-world objective.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c4af5f4f1fab" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Machine Learning Algorithms for “Classification”]]></title>
            <link>https://medium.com/@premkumar.jones/machine-learning-algorithms-for-classification-22e3deb3a099?source=rss-aa72d89a5ceb------2</link>
            <guid isPermaLink="false">https://medium.com/p/22e3deb3a099</guid>
            <dc:creator><![CDATA[Dr. Premkumar Borugadda, M.Tech.,Ph.D., Postdoc]]></dc:creator>
            <pubDate>Fri, 05 Jan 2024 02:10:26 GMT</pubDate>
            <atom:updated>2024-01-05T02:11:13.808Z</atom:updated>
            <content:encoded><![CDATA[<h3>Classification of Machine Learning Algorithms for “Classification”</h3><p>What is a classification?</p><p>Examples of Classification Problems</p><p>Types of classification tasks</p><p>Examples of Classification Problems</p><p><strong>What is a classification?</strong></p><p>Classification is a technique to categorise our data into distinct classes, where we can assign labels to each class.</p><p>Classification is a field of research that uses machine learning techniques to classify objects, things, images, and text.</p><ul><li>Classification in machine learning is a kind of supervised learning approach in which the ML algorithms are applied to the data and create a classification model that is used to classify the new data for accurate prediction.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4bpe6lGzOaD6MXXfjH-7yA.png" /><figcaption>Classification Process</figcaption></figure><p><strong>Examples of Classification Problems</strong></p><p>• Let&#39;s say you have your shop and you want to figure out if one of your customers is visiting it again or not.</p><p>• The answer to that question can only be a ‘Yes’ or ‘No’.</p><p>These kinds of problems in machine learning are known as classification problems.</p><p>•Classification problems normally have a <em>categorical output </em>like</p><p>• ‘yes’ or ‘no’</p><p>• ‘1’ or ‘0’</p><ul><li>‘True’ or ‘false’.</li></ul><p><strong>Types of classification tasks</strong></p><p>• Two types of classification tasks</p><p>1. Binary Classification</p><p>2. Multi-Class Classification</p><ol><li><strong>Binary Classification</strong></li></ol><p>• Binary classification with only two distinct classes or with two possible outcomes</p><p>•Examples:</p><p>• Male and Female</p><p>• Email spam detection (spam or not).</p><ul><li>Disease prediction (yes or no).</li></ul><p>• Popular algorithms that can be used for <em>binary classification </em>include:</p><p>1. Logistic Regression</p><p>2. K-Nearest Neighbors (KNN)</p><p>3. Decision Trees</p><p>4. Support Vector Machine (SVM)</p><p>5. Naive Bayes</p><p><strong>2. Multi-Class Classification</strong></p><p>• Multi-class classification refers to those classification tasks that have <em>more than two class labels</em>.</p><p>classification with more than two distinct classes.</p><p>1. classification of types of soil</p><p>2. classification of types of crops</p><p>• Popular algorithms that can be used for multi-class classification include:</p><p>•</p><p>1. k-Nearest Neighbors.</p><p>2. Decision trees.</p><p>3. Naive Bayes.</p><p>4. Random Forest.</p><p>5. Gradient Boosting.</p><p>● Ex. Wine quality</p><p><strong>Examples of Classification Problems</strong></p><p>1. Text categorization (e.g., spam filtering)</p><p>2. Fraud detection</p><p>3. Machine vision (e.g., face detection)</p><p>4. Natural-language processing</p><p>5. Market segmentation</p><p>6. Bioinformatics</p><p>7. Bank customer&#39;s loan payment willingness prediction.</p><p>8. Cancer tumour cell identification.</p><p>9. Sentiment analysis</p><p>10. Drugs classification</p><p><strong>Evaluation of a Classification Model</strong></p><p>•In machine learning, once we have a result of the classification problem, how do we measure how accurate our classification is?</p><p>•Metrics In a <em>classification problem</em>, the model performance is measured using the <strong><em>confusion matrix</em></strong>.</p><p>i.e., how accurately the true positives and true negatives were predicted.</p><p>• The different evaluation metrics are:</p><p>1. Accuracy</p><p>2. Misclassification Rate (Error Rate )</p><p>3. Precision</p><p>4. Recall/Sensitivity</p><p>5.F1 Score</p><p>6. Specificity</p><p>7. ROC (Receiver Operator Characteristic)</p><p>8.AUC( Area Under the Curve)</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=22e3deb3a099" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Roadmap for Mathematics in ‘Machine Learning’]]></title>
            <link>https://medium.com/@premkumar.jones/roadmap-of-mathematics-for-machine-learning-c229ed2021cd?source=rss-aa72d89a5ceb------2</link>
            <guid isPermaLink="false">https://medium.com/p/c229ed2021cd</guid>
            <dc:creator><![CDATA[Dr. Premkumar Borugadda, M.Tech.,Ph.D., Postdoc]]></dc:creator>
            <pubDate>Sat, 28 Oct 2023 05:55:02 GMT</pubDate>
            <atom:updated>2023-10-28T06:12:25.351Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*UWg9rPCmCg8WfoQjbu6lMw.jpeg" /></figure><p>Mathematics plays a foundational role in machine learning (ML).</p><p>It’s a skill to understand the math behind machine learning systems.</p><p>Machine learning and deep learning are built upon three pillars.</p><p>1. Linear Algebra</p><p>2. Calculus</p><p>3. Probability and Statistics</p><h3><strong>1. Linear Algebra</strong></h3><p>Linear algebra is a branch of mathematics that focuses on studying linear equations and functions, as well as their mathematical representation via matrices, vectors, tensors, and other related mathematical structures.</p><p>• In Machine Learning algorithms, data is often represented as matrices.</p><p>Matrix operations are used to manipulate data.</p><p>•So it is very important to have a basic knowledge of Linear Algebra to understand and make ML algorithms.</p><ul><li>Linear algebra is a fundamental mathematical field with various applications in machine learning and deep learning.</li></ul><p><strong>Examples of linear algebra concepts and their applications in these fields</strong></p><p>1. Vectors and Matrices</p><p>2. Matrix Operations (Addition, Multiplication, and Inversion)</p><p>3.Eigenvalues and Eigenvectors</p><p>4. Singular Value Decomposition (SVD)</p><p>5. Norms and Metrics</p><p><strong>Vector Space</strong></p><p>•Vector spaces are fundamental in machine learning as they provide a structured way to manipulate data, a mathematical framework for representing data and features.</p><p>•A vector space, also known as a linear space, is a fundamental concept in linear algebra.</p><p>Here are some examples of how vector spaces are used in machine learning.</p><ol><li><strong>Feature Vectors</strong></li></ol><p>Feature vectors are a fundamental concept in machine learning, where data is represented as numerical vectors with features (attributes or variables) that describe the data points.</p><p>Each element in a feature vector corresponds to a particular feature, and these vectors are used as inputs to machine learning models.</p><p>Here are some examples of feature vectors in different domains of machine learning.</p><p>1. Image Classification</p><p>2. Text Classification</p><p>3. Speech Recognition</p><p>4. Time Series Analysis</p><p>5. Healthcare</p><p>6. Anomaly Detection</p><ol><li><strong>Image classification</strong></li></ol><p>•In image classification, each image is represented as a high-dimensional vector where each dimension corresponds to a pixel’s intensity value.</p><p>•In computer vision, images can be represented as vectors of visual features extracted from different layers of a deep convolutional neural network.</p><p><strong>Application</strong>:</p><p>Feature vectors are used to represent data points, and machine learning models operate on these vectors to make predictions or classifications.</p><p>These feature vectors are used for tasks such as object detection, image classification, and image generation.</p><p>2. <strong>Text Classification</strong></p><p>• In natural language processing (NLP), documents or text segments can be represented as feature vectors using techniques like TF-IDF (Term Frequency-Inverse Document Frequency) or word embeddings (e.g., Word2Vec).</p><ul><li>Each word or term in the document is assigned a value, and the vector represents the combination of these values.</li></ul><p><strong>3. Speech Recognition</strong></p><ul><li>In speech recognition, audio signals are often transformed into feature vectors using methods like Mel-frequency cepstral coefficients (MFCCs).</li><li>Each feature in the vector may represent the energy or spectral content of the signal at a specific time frame.</li></ul><p><strong>4. Time Series Analysis</strong></p><ul><li>For time series forecasting, historical data points are often transformed into feature vectors that include lagged values, rolling statistics, and other time-related features.</li><li>For instance, to predict stock prices, a feature vector might include past stock prices and trading volumes.</li></ul><p><strong>5. Healthcare</strong></p><p>• In medical diagnosis, patient data can be represented as feature vectors containing various health-related attributes, such as age, blood pressure, cholesterol levels, and symptoms.</p><p>These vectors are used to predict disease outcomes or assess risk factors.</p><p><strong>6. Anomaly Detection</strong></p><ul><li>In cybersecurity, network traffic data can be represented as feature vectors, with each feature describing different aspects of the traffic, such as packet sizes, protocols, and source/destination IP addresses. Anomaly detection models use these vectors to identify unusual network behaviour.</li></ul><p><strong>Matrices and their operations</strong></p><ul><li>Matrices and their operations are fundamental tools in various machine learning algorithms and data analysis ranging from linear regression and dimensionality reduction techniques like PCA to deep learning with neural networks, as they are used to represent and manipulate data, features, and models.</li></ul><p>•Here are some common matrix operations in machine learning with examples:</p><p><strong>1. Matrix Representation of Data</strong>:</p><p>• Tabular data, such as datasets of numerical values with rows representing observations and columns representing features.</p><p>Each row can be seen as a data point, and the entire dataset is represented as a matrix.</p><p><strong>2. Matrix Transposition</strong>:</p><p>• Transposing a matrix swaps its rows and columns.</p><ul><li>Example: When working with tabular data, transposing a dataset may be useful to change the orientation of data points and features.</li></ul><p><strong>3. Matrix Multiplication (Dot Product)</strong>:</p><p>•Multiplying two matrices, where the number of columns in the first matrix must be equal to the number of rows in the second matrix.</p><p>•<strong>Example</strong>: In neural networks, matrix multiplication is used to calculate activations in each layer by multiplying the input matrix (data or activations from the previous layer) with a weight matrix.</p><p><strong>4. Matrix-Vector Multiplication</strong>:</p><p>•Multiplying a matrix by a vector (column matrix), resulting in a new vector.</p><ul><li><strong>Example</strong>: In linear regression, you can represent the coefficients as a column vector and the features as a matrix, and the product of these two yields the predicted values.</li></ul><p><strong>Determinants</strong></p><p>•Determinants are mathematical concepts that have applications in various areas of mathematics, including linear algebra, which plays a crucial role in machine learning.</p><p>•In machine learning, determinants are primarily used in contexts related to linear transformations, linear independence, and the properties of matrices.</p><p>•Examples of how determinants are used in machine learning.</p><p><strong>Matrix Rank</strong>:</p><ul><li>Example: Determinants can be used to compute the rank of a matrix. In machine learning, when dealing with feature selection or feature engineering, it’s essential to know the rank of a feature matrix to determine linear independence among features.</li></ul><p><strong>Singular Value Decomposition (SVD)</strong></p><p>•Singular Value Decomposition (SVD) is a fundamental matrix factorization technique used in linear algebra and various areas of data analysis, including machine learning.</p><ul><li><strong>Example</strong>: In dimensionality reduction techniques like Principal Component Analysis (PCA), the determinant of the covariance matrix can be used to calculate the proportion of total variance retained in the data when reducing dimensions.</li></ul><p><strong>Applications of SVD in Machine Learning</strong></p><p>• Dimensionality<strong> Reduction: </strong>SVD is used in techniques like Principal component analysis (PCA) to reduce the dimensionality of data while preserving as much information as possible.</p><p>• Linear<strong> Regression</strong>: SVD can help solve linear regression problems, especially when dealing with multicollinearity among features.</p><ul><li>Image<strong> Processing</strong>: SVD is used in various image processing tasks, including denoising, deblurring, and feature extraction.</li></ul><p><strong>4. Matrix Eigenvalue and Eigenvector Calculation</strong>:</p><p>•Finding the eigenvalues and eigenvectors of a matrix.</p><ul><li><strong>Example</strong>: In Principal Component Analysis (PCA), eigen values and eigen vectors of the covariance matrix are used to determine the principal components.</li></ul><p>2 — Statistics</p><p><em>Statistics plays a crucial role in various aspects of machine learning, from data preprocessing and exploration to model evaluation and interpretation.</em></p><p><em>Here are some key statistical concepts and their examples in machine learning:</em></p><p><em>Two Types of Statistics</em></p><p><em>2.1 Descriptive Statistics</em></p><p><em>2.2 Inferential Statistics</em></p><p><strong><em>2.1 Descriptive Statistics</em></strong></p><p>Descriptive statistics are essential for understanding the characteristics of data before building machine learning models.</p><p>• They help identify outliers, assess data quality, and gain initial insights into patterns and distributions, which can guide subsequent data preprocessing and modelling steps.</p><ul><li>Descriptive statistics in machine learning are used to summarize and describe the main characteristics of a dataset. They provide insights into the data’s central tendency, variability, and distribution.</li></ul><p>•Here are some common descriptive statistics and their examples in machine learning:</p><p><strong>1. Measures of Central Tendency</strong>:</p><p>•<strong>Mean</strong>: It is the average value of a dataset.</p><p>•<strong>Median</strong>: It is the middle value when the data is sorted, or the value that separates the higher half from the lower half.</p><p>•<strong>Mode</strong>: It is the most frequently occurring value in a dataset.</p><ul><li><strong>Example</strong>: In a dataset of house prices, you can calculate the mean, median, and mode to understand the price and its distribution.</li></ul><p>•<strong>Quartiles</strong>: Quartiles divide the data into four parts (Q1, Q2, Q3), where Q2 is the median.</p><p><strong>Frequency Distribution</strong>:</p><p>•<strong>Histograms</strong>: Histograms show the distribution of data by dividing it into bins and counting the frequency of values in each bin.</p><p>•<strong>Frequency Tables</strong>: These tables list the values in a dataset along with their corresponding frequencies.</p><ul><li><strong>Example</strong>: Creating a histogram of customer ages can help visualize the age distribution of customers in a retail dataset.</li></ul><p>• Skewness<strong> and Kurtosis:</strong></p><p>• Skewness: It measures the asymmetry of the data distribution. Positive skewness indicates a right-skewed distribution, while negative skewness indicates a left-skewed distribution.</p><p>• Kurtosis: Kurtosis is a statistical measure that quantifies the distribution of data points in a dataset.</p><p>• It provides information about the shape of the distribution, specifically how the data points are distributed in the tails relative to the centre of the distribution.</p><ul><li>Example: Kurtosis can be useful in various fields, including finance, statistics, and data analysis, to assess the nature of data distributions.</li></ul><p><strong><em>2.1 Descriptive Statistics</em></strong></p><p>2.1.1 Variables</p><p>2.1.2 Mean</p><p>2.1.3 Median</p><p>2.1.4 Mode</p><p>2.1.5 Standard Deviation</p><p>2.1.6 Variance</p><p>2.1.7 Range</p><p>2.1.8 Percentile</p><p>2.1.9 Skewness</p><p>2.1.10 Kurtosis</p><p><strong><em>2.2 Inferential Statistics</em></strong></p><p>Inferential statistics in machine learning involve making predictions, and drawing conclusions.</p><p>•It’s a crucial part of machine learning for tasks such as hypothesis testing, statistical modelling, and generalization.</p><p>Here are some examples of inferential statistics in machine learning:</p><p>2.2.1 Sampling Distributions</p><p>2.2.2 Central Limit Theorem</p><p>2.2.3 Hypothesis Testing</p><p>2.2.4 Confidence Intervals</p><p>2.2.5 T-Tests</p><p>2.2.6 Analysis of Variance (ANOVA)</p><p>2.2.7 Chi-Square Test</p><p>2.2.8 Regression Analysis</p><p>2.2.9 Bayesian Inference</p><p><strong>3 — Probability</strong></p><p>•Probability is a powerful tool and fundamental concept in machine learning for modeling and decision-making under uncertainty.</p><p>•It enables the development of probabilistic models, Bayesian inference, probabilistic programming, and more, which are essential for tackling various real-world problems in machine learning and artificial intelligence.</p><p>3.1 Probability Distribution</p><p>3.2 Conditional Probability</p><p>3.3 Bayes’ Theorem</p><p><strong>4 — Objective Functions</strong></p><ol><li>Mean Squared Error (MSE): Measures the average squared difference between predicted and actual values.</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/351/1*EZufwN5Q63NUc4dL5gSTQg.png" /></figure><p>2. Mean Absolute Error (MAE): Measures the average absolute difference between predicted and actual values.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/375/1*ebwZKnRkBFqAzn9LFe4phw.png" /></figure><p>3. Huber Loss: A combination of MSE and MAE, providing robustness to outliers.</p><p><strong>Classification</strong>:</p><p>1. <strong>Cross-Entropy Loss </strong>(Log Loss): Measures the dissimilarity between predicted class probabilities and actual class labels in binary or multi-class classification.</p><p>2. <strong>Hinge Loss</strong>: Used in Support Vector Machines (SVMs) for binary classification.</p><p>3. <strong>Categorical Cross-Entropy</strong>: Used for multi-class classification when class labels are one-hot encoded.</p><p>4. <strong>Sparse Categorical Cross-Entropy</strong>: Used for multi-class classification when class labels are integers.</p><p>5. <strong>Gini Impurity</strong>: Used in decision tree algorithms, such as Gini-based impurity for CART (Classification and Regression Trees).</p><p><strong>Clustering</strong>:</p><p>•<em>K-Means Loss</em>: Measures the distance between data points and cluster centroids.</p><p>•<em>Gaussian Mixture Model (GMM) Loss</em>: Measures the likelihood of data points under a Gaussian mixture model.</p><p>•<em>Spectral Clustering Loss</em>: Measures the similarity between data points in a graph-based clustering approach.</p><ul><li><em>DBSCAN Loss</em>: Measures the density-based clustering quality.</li><li>•The choice of objective function depends on the nature of the problem and the characteristics of the data.</li><li>•Different tasks and algorithms may require different loss functions to optimize model parameters effectively and achieve the desired performance.</li></ul><p><strong>5- Regularization</strong></p><p>•Regularization is a technique in machine learning that is used to prevent overfitting and improve the generalization of models.</p><p>•It does by adding a penalty term to the model’s loss function.</p><ul><li>The choice of regularization method depends on the specific problem, the nature of the data, and the algorithm being used.</li><li><strong>L1 Regularization (Lasso)</strong></li><li>Objective: Encourages sparsity in feature selection by adding the absolute values of the model’s coefficients to the loss function.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/1*9B45s1LexPagcPkyp075iQ.png" /></figure><p><strong>Example</strong>:</p><ul><li>In linear regression, L1 regularization can be used to select a subset of the most important features. It sets some coefficients to zero, effectively performing feature selection.</li></ul><p><strong>5- Regularization</strong></p><p>5.1. L1 Regularization (Lasso Regression)</p><p>5.2. L2 Regularization (Ridge Regression)</p><p>5.3. Elastic Net Regularization</p><p>5.4. Dropout Regularization</p><p>5.5. Data Augmentation</p><p>5.6. Early Stopping</p><p><strong>6-Distribution</strong></p><p>6.1 Bernoulli Distribution</p><p>6.2 Binomial Distribution</p><p>6.3 Multinomial Distribution</p><p>6.4 Normal (Gaussian) Distribution</p><p>6.5 Uniform Distribution</p><p>6.6 Exponential Distribution</p><p>6.7 Poisson Distribution</p><p><strong>7- Calculus</strong></p><p>1. Functions</p><p>2. Derivatives</p><p>3. Product and Chain Rule Differentiation</p><p>4. Optimisation (e.g., gradient descent)</p><p>5. Composite functions</p><p>6. Partial Derivatives</p><p>7. Higher-order derivatives</p><p>8. Integrals</p><p>9.Limits</p><p><strong>Calculus</strong></p><p><strong>Example:</strong></p><p>•Gradient Descent is a fundamental optimization algorithm used in training machine learning models.</p><ul><li>Calculus is used to compute gradients to update model parameters iteratively.</li></ul><p><strong>Note: For more details, watch this video</strong></p><p><a href="https://www.youtube.com/watch?v=4D-OYFpAaxw&amp;t=54s">https://www.youtube.com/watch?v=4D-OYFpAaxw&amp;t=54s</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c229ed2021cd" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Gradient Descent Algorithm in Machine Learning]]></title>
            <link>https://medium.com/@premkumar.jones/gradient-descent-algorithm-in-machine-learning-f43556e32d48?source=rss-aa72d89a5ceb------2</link>
            <guid isPermaLink="false">https://medium.com/p/f43556e32d48</guid>
            <dc:creator><![CDATA[Dr. Premkumar Borugadda, M.Tech.,Ph.D., Postdoc]]></dc:creator>
            <pubDate>Wed, 20 Sep 2023 04:23:23 GMT</pubDate>
            <atom:updated>2023-09-20T04:25:39.064Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5ME4PIztngXNlZYXO7gpZQ.jpeg" /><figcaption>Gradient Descent Algorithm</figcaption></figure><p>• GDA helps us<strong> find the best fit line </strong>in a very less number of iterations in an efficient way.</p><p>i.e., GDA is an iterative optimisation algorithm to find the minimum cost function.</p><p>• Let&#39;s apply the GDA to find the ‘m’ &amp; ‘b’ values.</p><p>• Initially, we have to take m and b values as zeros.</p><p>•‘L’ is a <strong><em>learning rate </em></strong>or <strong><em>alpha rate </em></strong>which controls the “<em>how much the value of ‘m’ changes in each step</em>”.</p><p>• i.e., ‘L’ determines the <em>size of steps</em> to reach a global minimum.</p><ul><li>‘L’ could be a small value like 0.00001 for good accuracy.</li><li>• Let&#39;s suppose, epochs (iterations) are 100, which means the number of times to perform the gradient descent is 100 times.</li><li>which means, epochs refer to the number of times we train our model to find the best-fit values like ‘m’ and ‘b’.</li><li>• To find the best-fit line for all given data points (n).</li><li>•So, our predictions are much more accurate.</li><li>•MSE= 1/n∑_(i=1)^n▒(yi-y_pred) 2.</li><li>•Gradient Descent to reduce the error with the “derivation function” and “learning rate(L)” or “Alpha rate”.</li></ul><p><strong>How to Minimise the Error</strong></p><p>• The machine always tries to find the best-fitted values like ‘m’ &amp; ‘b’ for the best-fit line at any point in time.</p><p>• Suppose we have slope (m), intercept (b), and error (e), try to find the relation among the three.</p><p>A gradient descent graph is a 3D graph.</p><p>• ML is gradient descent in the case of linear regression.</p><p>• ‘m’ and ‘b’ are the only factors that derive the relationship between ‘X’ and y.</p><p>• Error can be calculated as = (yi-y_pred)</p><p>• To get the best-fit line, we must reduce the error (e) by applying the cost function through the gradient descent algorithm.</p><p><strong>Cost Function or Loss Function:</strong></p><p>• The cost function tells us “<strong><em>how good our model is</em></strong>” and <em>makes predictions</em> for a given value of ‘m’ and ‘b’.</p><p>The cost function evaluates the <em>performance</em> of our machine-learning algorithm.</p><ul><li>The slope of this curve tells us how to update our parameters to make the model more accurate.</li><li><strong>Minimising the Cost Function</strong></li><li>• The goal of any Machine Learning Algorithm is to minimise the Cost Function.</li><li>• In total, there are ’n’ points in the given dataset, for all those ’n’ points, we have to minimise the error (e) with the help of the cost function.</li><li>• Cost function or MSE=1/n∑_(i=1)^n▒(yi-y_pred) 2.</li><li>• How<strong> do we actually minimise the Cost function?</strong></li><li>• Using Gradient Descent minimises the cost function.</li><li>• Now we have to find ‘m’ &amp; ‘b’ values for the best-fit line by using Gradient descent.</li></ul><p>For more details, watch the video</p><p><a href="https://www.youtube.com/watch?v=4n1Tjkh7g0A">https://www.youtube.com/watch?v=4n1Tjkh7g0A</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f43556e32d48" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Linear Regression in Python]]></title>
            <link>https://medium.com/@premkumar.jones/linear-regression-cfe04e09488c?source=rss-aa72d89a5ceb------2</link>
            <guid isPermaLink="false">https://medium.com/p/cfe04e09488c</guid>
            <category><![CDATA[linear-regression]]></category>
            <category><![CDATA[simple-linear-regression]]></category>
            <dc:creator><![CDATA[Dr. Premkumar Borugadda, M.Tech.,Ph.D., Postdoc]]></dc:creator>
            <pubDate>Tue, 19 Sep 2023 06:36:24 GMT</pubDate>
            <atom:updated>2023-09-19T07:22:05.201Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>What is Regression Analysis?</strong></p><p>•Regression in statistics is predicting a Label(or Dependent Variable) based on the features(Independent Variables).</p><p>•Regression analysis is a form of predictive modelling technique which investigates the relationship between a <strong>dependent </strong>(target) and <strong>independent variable (s)</strong> (predictor).</p><p>•Regression analysis is a crucial tool for <em>analyzing</em> <em>data</em>.</p><p>•Here, we fit a curve/line to the data points so that the differences between the distances of data points from the curve or line are minimized.</p><ul><li>This technique is used for forecasting and finding the <a href="https://www.analyticsvidhya.com/blog/2015/06/establish-causality-events/">relationship</a> between the variables.</li></ul><p><strong>The use of Regression</strong></p><p>•Regression analyses <em>predict the relationship between two or more features</em>.</p><p>•The benefits of using Regression analysis are as follows:</p><p>•It shows the significant relationships between the Label (dependent variable) and the features(independent variable).</p><p>•It indicates the strength of the impact of multiple independent variables on a dependent variable.</p><ul><li>These benefits help market researchers/data analysts/data scientists eliminate and evaluate the best set of variables to be used for building predictive models.</li></ul><p><strong>Linear Regression</strong></p><p>•Regression Models are used <em>to predict the continuous values</em>.</p><p>•It is a supervised technique.</p><p>•Linear Regression is a <em>statistical analysis</em> for <em>predicting the value of a quantitative variable.</em></p><p>•Linear Regression is one of the most fundamental algorithms in Machine Learning (ML).</p><p>-In this technique, the dependent variable is continuous, independent variable(s) can be continuous or discrete, and the nature of the regression line is linear.</p><ul><li>Linear Regression establishes a relationship between the dependent variable (Y) and one or more independent variables (X) using a Best fit straight line (also known as a <strong>regression line</strong>).</li></ul><p><strong>The building blocks of a Linear Regression Model are:</strong></p><p>•Discreet/continuous independent variables.</p><p>•A best-fit regression line.</p><p>•Continuous dependent variable.</p><p>• A Linear Regression model predicts the dependent variable using a regression line based on the independent variables. The equation of the Linear Regression is:</p><p>• Y=m*X +b + e</p><p>•Where ‘<strong>b</strong>’ is the intercept, ‘<strong>m</strong>’ is the slope of the line, and ‘<strong>e</strong>’ is the error term.</p><ul><li>The above equation is used to predict the value of the target variable based on the given predictor variable(s).</li><li>•The difference between simple linear regression and multiple linear regression is that multiple linear regression has more than one (&gt;1) independent variable, whereas simple linear regression has only one independent variable.</li></ul><p><strong>Simple Linear Regression(SLR)</strong></p><p>•Simple Linear regression is a method for predicting a <strong>quantitative response</strong> using a <strong>single feature</strong> (“input variable”).</p><p>•The mathematical equation is:</p><p>Y=m*X +b</p><p>• Where ‘y’ is the Response or Target variable.</p><p>• ‘X’ is the independent (input) feature</p><p>• ‘m’ is the coefficient of ‘X’ or slope</p><p>• ‘b’ is the intercept</p><p>•Here, m &amp; b are the <strong>model coefficients</strong>.</p><p>•To create a model, we must “learn” the values of these coefficients.</p><ul><li>Once we have the values of these coefficients, we can use the model to predict the target variable.</li></ul><p><strong>Evaluation metrics of linear regressio<em>n</em></strong></p><p>• We can compare the actual and predicted values to calculate the accuracy of our regression model.</p><p>Evaluation metrics play a key role in the development of a model.</p><p>• The number of model evaluation metrics are MAE, MSE, RMSE and R-Squaure.</p><ul><li>Error: the measure of how far the data is from the fitted regression line.</li></ul><p><strong><em>MAE (Mean Absolute Error)</em></strong></p><ul><li>MAE is the mean of the absolute value of the errors. This is the easiest of the metrics to understand.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/610/1*H-VJkO91OrcUeCAIcx4xVA.png" /></figure><ul><li>where ‘yi’ is the Actual output, ‘yi hat’ is the predicted output and ’N’ number of samples in the dataset.</li></ul><p><em>MSE (Mean Square Error)</em></p><p>• MSE is the mean of the squared error.</p><p>• It&#39;s more popular than MAE because the focus is on large errors.</p><ul><li>Due to the exponential increase in larger errors in comparison to smaller errors.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/749/1*drzeQ62YZmdVc8XbAJYpZQ.png" /></figure><ul><li><strong><em>RMSE (Root Mean Square Error)</em></strong></li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/674/1*g7fwzY8ZSZNBOa7_Lu1eFA.png" /></figure><ul><li>RMSE is the square root of the mean squared error.</li><li>This is one of the most popular of the evolution metrics.</li></ul><p><strong><em>R-Squared</em></strong></p><p>• It is a popular metric for the accuracy of our model.</p><p>It represents how close the data values are to the fitted regression line.</p><p>The higher the R-square, the better the model fits our data.</p><ul><li>If R-square is negative it means that the model is worse than predicting the mean.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DQ98-J5SfkneEpSsmzrf4A.png" /></figure><p>For Full information about this linear regression, watch this video in my YouTube channel.</p><p><a href="https://www.youtube.com/watch?v=4zq_gw2zeRE&amp;t=806s">https://www.youtube.com/watch?v=4zq_gw2zeRE&amp;t=806s</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=cfe04e09488c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Types of Variables in the Data]]></title>
            <link>https://medium.com/@premkumar.jones/types-of-variables-in-the-data-48b1fcbf1a1b?source=rss-aa72d89a5ceb------2</link>
            <guid isPermaLink="false">https://medium.com/p/48b1fcbf1a1b</guid>
            <dc:creator><![CDATA[Dr. Premkumar Borugadda, M.Tech.,Ph.D., Postdoc]]></dc:creator>
            <pubDate>Tue, 30 May 2023 01:45:20 GMT</pubDate>
            <atom:updated>2023-05-30T02:10:02.959Z</atom:updated>
            <content:encoded><![CDATA[<h3>variables in a data set</h3><p>A variable is a property that can be evaluated and can take on different amounts. Variables include size, salary, country, gender, college grades, and other variables. Variables can be put into two main groups: those that are categorical and those that are numbers. Then, each group is split into two subcategories: nominal or ordinal for categorical variables and discrete or continuous for numeric variables. In this part, we’ll talk quickly about these types.</p><p>•A variable is any characteristic, number, or quantity that can be measured or counted.</p><p>•Age (21, 35, 62, …)</p><p>•Gender (male, female)</p><p>•Income (Rs. 20000, Rs. 35000, Rs. 45000, …)</p><p>•House price (Rs. 350000, Rs. 570000, …)</p><p>•Country of birth (China, Russia, Costa Rica, …)</p><p>•Eye colour (brown, green, blue, …)</p><ul><li>Vehicle makes (Ford, Volkswagen, …)</li></ul><p>• Most variables in a data set can be classified into one of two major types:</p><p>•<strong>Numerical variables</strong></p><p>•<strong>Categorical variables</strong></p><p><strong>Numerical variables</strong></p><p>The values of a numerical variable are numbers. They can be further classified into discrete and continuous variables.</p><p><strong>Discrete numerical variable</strong></p><p>• A variable whose values are whole numbers (counts) is called discrete.</p><p>The following are examples of discrete variables:</p><p>• Number of children in the family</p><ul><li>Number of active bank accounts</li></ul><h3><strong>Continuous Numerical variable</strong></h3><p>A variable that may contain any value within some range is called continuous.</p><p>• For example, the total amount paid by a customer in a supermarket is continuous.</p><p>• The customer can pay Rs. 20.5, Rs. 13.10, Rs. 83.20 and so on.</p><p>Other examples of continuous variables are:</p><p>•House price (it can take any value) (Rs. 350000, 57000, 1000000, …)</p><ul><li>Time spent surfing a website (3.4 seconds, 5.10 seconds, …)</li></ul><h3>Categorical variables</h3><p>• The values of a categorical variable are selected from a group of <strong>categories</strong>, also called <strong>labels</strong>.</p><p>Examples are gender (male or female) and marital status (never married, married, divorced, or widowed).</p><p>•Categorical variables can be further categorized into</p><p>•Ordinal Variable</p><p>Nominal variables</p><h3><strong>Ordinal categorical variables</strong></h3><p>• Categorical variables in which categories can be meaningfully ordered are called ordinal.</p><p>•For example:</p><p>• The student’s grade in an exam (A, B, C, or fail).</p><p>• Days of the week can be ordinal, with Monday = 1 and Sunday = 7.</p><ul><li>Educational level, with the categories elementary school, high school, college graduate, and PhD ranked from 1 to 4.</li></ul><h3><strong>Nominal categorical variable</strong></h3><p>• There isn’t an intrinsic order to the labels.</p><p>• For example, the country of birth (Argentina, England, or Germany) is nominal.</p><p>Other examples of nominal variables include:</p><p>•Postcode</p><p>•Vehicle make (Citroen, Peugeot, …)</p><p>• Sometimes categorical variables are coded as numbers when the data are recorded (e.g., gender may be coded as 0 for males and 1 for females).</p><p>The variable is still categorical.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=48b1fcbf1a1b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Introduction –’ Machine Learning]]></title>
            <link>https://medium.com/@premkumar.jones/introduction-machine-learning-cdf8fc1f7979?source=rss-aa72d89a5ceb------2</link>
            <guid isPermaLink="false">https://medium.com/p/cdf8fc1f7979</guid>
            <dc:creator><![CDATA[Dr. Premkumar Borugadda, M.Tech.,Ph.D., Postdoc]]></dc:creator>
            <pubDate>Fri, 26 May 2023 01:40:46 GMT</pubDate>
            <atom:updated>2023-09-19T06:44:48.857Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>Introduction –’ Machine Learning</strong></p><p>• Learning is defined as the acquisition of knowledge or skills through experience and study.</p><p>Machine learning is still a fairly new concept. We can teach machines how to learn, and some machines can even learn on their own. This magical phenomenon is called machine learning.</p><p>Machine learning is a subfield of artificial intelligence. Its goal is to enable computers to learn on their own.</p><ul><li>A machine’s learning algorithm enables it to identify patterns in observed data, build models that explain the world, and predict things without having explicit pre-programmed rules and models.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/653/1*66QSFVl6G3dc634vawQhSQ.png" /><figcaption>Source: <a href="https://vitalflux.com/"><strong>Data Analytics</strong></a><strong> course</strong></figcaption></figure><ul><li>Applications of Machine Learning</li><li>1. <strong>Healthcare</strong>: Predicting patient diagnostics for doctors to review</li><li>2. <strong>Social Network</strong>: Predicting certain match preferences on a dating website for better compatibility</li><li>3. <strong>Finance</strong>: Predicting fraudulent activity on a credit card</li><li>4. <strong>E-commerce</strong>: Predicting customer churn</li><li>5. <strong>Biology</strong>: Finding patterns in gene mutations that could represent cancer.</li></ul><p>“Most of the knowledge in the world in the future is going to be extracted by machines and will reside in machines&quot;—Yann<strong> LeCun</strong>, Director of AI Research, Facebook.</p><p>Machine Learning — Computer computer algorithms that can be trained for specific applications and used for future predictions.</p><p>Machine learning algorithms automatically build a mathematical model using sample data — also known as “training data&quot;—to make decisions without being specifically programmed to make those decisions.</p><p>Machine learning (ML) is the study of computer algorithms that improve automatically through experience.</p><p>•Machine learning algorithms build a model based on sample data, known as “training data”, in order to make predictions or decisions without being explicitly programmed to do so</p><p>• According to <strong>McKinsey</strong>, he defined machine learning as the: “algorithms that can learn from data without relying on rules-based programming”.</p><p>According to <strong>Arthur Samuel</strong>, “Machine learning is the ability of a computer to perform the task itself without being explicitly programmed.”</p><p>•<strong><em>Def (Mitchell 1997): </em></strong>A computer program is said to learn from experience <strong>E </strong>with respect to some class of tasks T and performance measure P, if its performance at tasks in T, as measured by P, improves by experience E.</p><p>•<strong><em>For example</em></strong>, a computer program that learns to play checkers might improve its performance as <strong><em>measured by its ability to win </em></strong>at the class of tasks involving <strong><em>playing checkers games, </em></strong>through experience <strong><em>obtained </em></strong>by <strong><em>playing games against itself. </em></strong>In general, to have a well-defined learning problem, we must identity these</p><p>•Three features:</p><p>•the class of tasks,</p><p>•the measure of performance to be improved, and</p><p>•the source of experience.</p><p>•<strong><em>A checkers learning problem:</em></strong></p><p>•Task T: playing checkers</p><p>•Performance measure <strong>P</strong>: percent of games won against opponents</p><p>•Training experience <strong>E</strong>: playing practice games against itself.</p><p>References:</p><ol><li><a href="https://codeandhack.com/introduction-to-machine-learning/">https://codeandhack.com/introduction-to-machine-learning/</a></li><li><a href="https://www.claoudml.com/blog/machine-learning-for-beginners">https://www.claoudml.com/blog/machine-learning-for-beginners</a></li><li><a href="https://www.dataversity.net/a-brief-history-of-machine-learning/">https://www.dataversity.net/a-brief-history-of-machine-learning/</a></li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=cdf8fc1f7979" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>