<?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 Ruqayya Nasir Iro on Medium]]></title>
        <description><![CDATA[Stories by Ruqayya Nasir Iro on Medium]]></description>
        <link>https://medium.com/@rukynas08?source=rss-f448c518842e------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*eo26mYHFpROJVM8-x-mfLw.jpeg</url>
            <title>Stories by Ruqayya Nasir Iro on Medium</title>
            <link>https://medium.com/@rukynas08?source=rss-f448c518842e------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 12:44:18 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@rukynas08/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[A research blog on measuring how prompt wording shapes LLM labels]]></title>
            <link>https://medium.com/@rukynas08/a-research-blog-on-measuring-how-prompt-wording-shapes-llm-labels-973f4aa5ceb1?source=rss-f448c518842e------2</link>
            <guid isPermaLink="false">https://medium.com/p/973f4aa5ceb1</guid>
            <dc:creator><![CDATA[Ruqayya Nasir Iro]]></dc:creator>
            <pubDate>Tue, 26 May 2026 12:32:12 GMT</pubDate>
            <atom:updated>2026-05-26T12:33:58.497Z</atom:updated>
            <content:encoded><![CDATA[<p>My research sits at the intersection of natural language processing and data annotation. Data annotation is the process of labelling data so that machine learning models have something to learn from. For years this has been done by people, often through platforms that hire annotators on a project basis. I have worked on several of these platforms myself, including Outlier and RWS, which is part of why this topic interests me.</p><p>LLMs are now being used for the same work. Gilardi, Alizadeh and Kubli (2023) show that ChatGPT can outperform crowd workers on several annotation tasks, and the practice is becoming common. My concern is that LLM outputs are sensitive to small changes in how the prompt is phrased (Sclar et al., 2024). If labels shift with phrasing, the datasets built from them are less reliable than they look.</p><h4>The research question</h4><p><strong><em>How much do small prompt variations change the labels an LLM produces in a data annotation task?</em></strong></p><p>If two annotators are given slightly different instructions, you would expect their labels to mostly agree. The same should be true of an LLM given two versions of the same prompt. The literature suggests this is not always the case, and the implications for downstream models are real.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*sEghS8VH2HRP6t7RORSRVA.png" /></figure><h4>How and why I plan to study it</h4><p>I will use a public sentiment classification dataset with gold labels. Sentiment is a good starting point because the task is well defined and there are established datasets to choose from. I plan to use an open-source model, either Llama 3 or Mistral, run through Hugging Face. Open-source models are a better fit than closed APIs here because the model can be pinned to a version and the results can be reproduced by others. I’ll create a few prompt variations. These will include alterations to wording, the style of instruction and the inclusion or omission of examples. I will run each variant on the same data sample and then measure agreement between prompt pairs using Cohen’s kappa (Cohen, 1960).</p><p>LLMs are increasingly used in place of human annotators, and the datasets they produce go on to train other models. If the labels shift based on how the prompt is worded, then the datasets are not as reliable as they appear. I have spent a lot of time as a human annotator. Annotation platforms invest heavily in instructions and quality checks for human annotators, and the same care is not always applied when an LLM does the same work. This research is my way of looking at that gap from the inside.</p><h3>Using Notion to Manage My Research Project</h3><p>A dissertation has a lot going on, and things get lost when you rely on memory and notes scattered across different places. I have decided to use Notion as my main project management space.</p><p>Notion combines note-taking, databases, task lists, and document writing in one place (Notion Labs, 2024). I have not used it in depth before, so I compared a few tools. Trello is great for managing tasks, but it’s not good for longer notes. OneNote is good for taking notes, but not as strong for tracking tasks. Obsidian is powerful for connecting notes together, but it takes more setup. Notion sits somewhere in the middle and can handle both notes and tasks without too much extra work. The main reason for picking it is that everything stays in one place. For this project I need to track tasks, store literature notes, log experiment results, and draft blog posts.</p><h4>The wider technical stack</h4><p>Notion is the front end, but the actual research work runs on a stack of free, open tools:</p><ul><li><strong>Python</strong> for the analysis scripts. This is the standard language for data science work and gives me access to the libraries below.</li><li><strong>Hugging Face Transformers</strong> for running the open-source model. The library handles model loading, tokenisation, and inference in a few lines of code.</li><li><strong>Scikit-learn</strong> for the kappa calculation. The cohen_kappa_score function is the standard implementation and avoids me writing it from scratch.</li><li><strong>pandas</strong> for organising the labels, prompts, and outputs into dataframes I can compare easily.</li><li><strong>GitHub</strong> for version control on the code. To help keep track of which version of a script produced a given result.</li><li><strong>Zotero</strong> for reference management. It plugs into Word and exports Harvard citations directly, which saves a lot of formatting time later.</li></ul><p>Notion is not built for heavy data work, so the actual outputs will live in Python notebooks and CSV files. Notion will hold the summary and the link, not the raw data. The clearest benefit of this whole setup is visibility and traceability. I should be able to open one tab in Notion and see what is due, what I have read, and what I have tested, with the code itself version-controlled on GitHub. The experiment log matters most. Without a dated record, results from, for example, week three cannot be compared honestly to results from week ten.</p><h3>How I Will Measure Label Consistency</h3><p>Percentage agreement is the obvious first thing to look at, but it can be misleading. With only two sentiment labels, two annotators who label at random would still agree around half the time if both labels are equally likely. Simple percentage agreement does not adjust for this chance agreement (Artstein and Poesio, 2008). I therefore need a reliability metric that tells me whether the annotators agree more often than we would expect by chance alone.</p><h4>Cohen’s kappa</h4><p>Cohen’s kappa measures agreement between two annotators on categorical items, adjusted for chance (Cohen, 1960). The value runs from minus one to one. A score of one means perfect agreement. Zero means agreement no better than chance. A commonly cited guide from Landis and Koch (1977) is that scores above 0.6 indicate substantial agreement and scores above 0.8 indicate near-perfect agreement.</p><p>The reason kappa is more informative than percentage agreement is that it factors in how often each label appears. If a dataset is skewed towards one label, two annotators can agree most of the time just by picking the common label.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*k2mBCe7n0uyCBXjCBiSWTg.png" /></figure><h4>How I will apply it</h4><p>In my project the two annotators are not people. They are two prompt variants given to the same model on the same data sample. I will run the model using prompt A, then again using prompt B, and compare the two label sets with Cohen’s kappa. I plan to run this for several prompt pairs (for example, a baseline against a reworded version or a zero-shot prompt against a few-shot version). Each pair gives me one kappa score, and comparing scores across pairs should show which types of prompt change have the biggest effect on consistency.</p><h4>Sample size</h4><p>I plan to use a fixed sample of 500 items, run under each prompt variant. This is large enough for kappa to be estimated stably (Sim and Wright, 2005) and small enough to be practical given runtime and cost. Each prompt variant will be run three times so I can also measure within-prompt variation, separating model randomness from the effect of the prompt itself.</p><h4>Success criteria</h4><p>I set explicit thresholds in advance so the analysis has clear outcomes rather than being interpreted after the fact:</p><ul><li><strong>A mean pairwise kappa above 0.80</strong> would suggest prompts have little practical effect on labels.</li><li><strong>A mean kappa between 0.60 and 0.80</strong> would suggest moderate sensitivity that researchers should flag when using LLM annotation.</li><li><strong>A mean kappa below 0.60</strong> would suggest serious sensitivity and a real problem for the practice of LLM-based annotation at scale.</li><li><strong>A difference of 0.15 or more between two prompt pairs</strong> would count as a meaningful effect of that specific kind of prompt change.</li></ul><h4>Limitations I want to flag early</h4><p>Kappa can give low scores when one label dominates, even if the raw agreement is high; this is the so-called kappa paradox (Feinstein and Cicchetti, 1990). I will therefore report percentage agreement alongside kappa for transparency. Cohen’s kappa is defined for two annotators. If I later want to compare three or more prompt variants at once, I would need a multi-rater coefficient such as Fleiss’ kappa or Krippendorff’s alpha (Krippendorff, 2018).</p><h3>Adapting My Research Approach to a Fast-Moving Field</h3><p>LLMs are released, updated, and deprecated on a timescale that does not match how academic research usually runs. A typical dissertation takes months. A frontier model can be replaced in weeks (Ziems et al., 2024).</p><p>A traditional plan assumes the tools will be the same at the end as at the start. For my project I cannot assume this. I am dealing with this in two ways. First, an open-source model rather than a closed API. Open-source models can be pinned to a specific version and stored locally. A closed API can change underneath you without warning (Chen, Zaharia and Zou, 2023). For each run, we log the model version, date, prompt text, and dataset sample. Without that, you can’t honestly compare results over weeks.</p><h4>Adapting the research methods</h4><p>The methods need tailoring too. Cohen’s kappa was developed for human annotators. It still works for LLM outputs, but the interpretation is different. With humans, low agreement usually points to unclear instructions. With an LLM, low agreement between two prompts points to something about the model’s behaviour, not the annotators’ judgement.</p><p>Replicability is harder than usual. Even with the same model, the same prompt, and the same data, outputs can vary because of sampling settings. I plan to address this by setting the model to a low temperature, fixing random seeds where possible, and running each prompt multiple times to capture within-prompt variation.</p><p>The literature is moving fast too, a paper from two years ago may already be out of date. I plan to weigh recent work more heavily than I would for a stable field, while still anchoring the project in older foundational work on annotation reliability. Cohen’s 1960 paper on kappa is not going anywhere. The applied LLM literature needs to be read with the publication date in mind.</p><h3>References</h3><ul><li>Artstein, R. and Poesio, M., 2008. Inter-coder agreement for computational linguistics. <em>Computational Linguistics</em>, 34(4).</li><li>Chen, L., Zaharia, M. and Zou, J., 2023. How is ChatGPT’s behavior changing over time? <em>arXiv preprint</em> arXiv:2307.09009. Available at: <a href="https://arxiv.org/abs/2307.09009">https://arxiv.org/abs/2307.09009</a> [Accessed 21 May 2026].</li><li>Cohen, J., 1960. A coefficient of agreement for nominal scales. <em>Educational and Psychological Measurement</em>, 20(1).</li><li>Feinstein, A.R. and Cicchetti, D.V., 1990. High agreement but low kappa: I. The problems of two paradoxes. <em>Journal of Clinical Epidemiology</em>, 43(6).</li><li>Gilardi, F., Alizadeh, M. and Kubli, M., 2023. ChatGPT outperforms crowd workers for text-annotation tasks. <em>Proceedings of the National Academy of Sciences</em>, 120(30), e2305016120.</li><li>Krippendorff, K., 2018. <em>Content analysis: An introduction to its methodology</em>. 4th ed. Thousand Oaks, CA: Sage.</li><li>Landis, J.R. and Koch, G.G., 1977. The measurement of observer agreement for categorical data. <em>Biometrics</em>, 33(1), pp.159–174.</li><li>Notion Labs, 2024. <em>Notion: One workspace. Every team.</em> Available at: <a href="https://www.notion.so/">https://www.notion.so</a> [Accessed 25 May 2026].</li><li>Sclar, M., Choi, Y., Tsvetkov, Y. and Suhr, A., 2024. Quantifying language models’ sensitivity to spurious features in prompt design. In: <em>ICLR 2024</em>. Available at: <a href="https://arxiv.org/abs/2310.11324">https://arxiv.org/abs/2310.11324</a> [Accessed 26 May 2026].</li><li>Sim, J. and Wright, C.C., 2005. The kappa statistic in reliability studies: Use, interpretation, and sample size requirements. <em>Physical Therapy</em>, 85(3).</li><li>Ziems, C., Held, W., Shaikh, O., Chen, J., Zhang, Z. and Yang, D., 2024. Can large language models transform computational social science? <em>Computational Linguistics</em>, 50(1).</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=973f4aa5ceb1" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Role of Regression in Machine Learning]]></title>
            <link>https://medium.com/@rukynas08/the-role-of-regression-in-machine-learning-0e31bec04916?source=rss-f448c518842e------2</link>
            <guid isPermaLink="false">https://medium.com/p/0e31bec04916</guid>
            <dc:creator><![CDATA[Ruqayya Nasir Iro]]></dc:creator>
            <pubDate>Fri, 29 Mar 2024 22:12:00 GMT</pubDate>
            <atom:updated>2024-03-29T22:12:00.167Z</atom:updated>
            <content:encoded><![CDATA[<p>In machine learning, regression is similar to cutting a line through a graph’s point cloud. Assume you have a collection of points that correspond to various types of data, such as house prices and sizes. In order to make predictions about future data, regression assists us in determining which straight line best fits these points. For example, if we know the size of a house, regression can help us predict its price.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*XolBuOC16ihcTGfnrp1_qg.jpeg" /></figure><p>Regression is used in a wide range of applications nowadays. One common example is predicting house prices. Regression analysis is a tool used by real estate agents to examine information about past home sales, including the property’s size, number of bedrooms, and location. They can determine the price of newly constructed homes on the market by fitting a regression line to this data.</p><p>Another example is in finance. Stock market analysts use regression to predict the future price of stocks based on historical data. Regression models analyze a variety of variables, including economic indicators, market trends, and company earnings, to assist investors in making well-informed decisions when purchasing or disposing of stocks.</p><p>Regression analysis is used in healthcare to forecast patient outcomes. In order to determine the likelihood of specific diseases or conditions, physicians and researchers examine patient data, including age, medical history, and test results. Identifying diseases early and administering the right care can be greatly aided by this knowledge.</p><p>In marketing, regression helps businesses understand customer behavior. Businesses are able to forecast future sales trends and adjust their marketing strategies by examining data on past purchases, website visits, and demographic data.</p><p>Regression is an effective machine learning tool that, when used properly, can help us comprehend the relationships between variables and forecast future events. Its many uses in a variety of sectors spur innovation and decision-making in the contemporary world.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0e31bec04916" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why Clean Data is Key to Reliable Insights]]></title>
            <link>https://medium.com/@rukynas08/why-clean-data-is-key-to-reliable-insights-7a4347d1d453?source=rss-f448c518842e------2</link>
            <guid isPermaLink="false">https://medium.com/p/7a4347d1d453</guid>
            <dc:creator><![CDATA[Ruqayya Nasir Iro]]></dc:creator>
            <pubDate>Tue, 27 Feb 2024 16:12:07 GMT</pubDate>
            <atom:updated>2024-02-27T16:12:07.540Z</atom:updated>
            <content:encoded><![CDATA[<p>In today’s digital age, where data is everywhere, there’s a crucial step that often goes overlooked: data preparation. We will explore how data preparation can prevent data disasters by understanding why it matters.</p><p><strong>What is Data Preparation?</strong></p><p>Imagine you’re baking a cake. You wouldn’t just throw ingredients into the oven and hope for the best, right? Like preparing those ingredients, data preparation involves cleaning, organizing, and transforming raw data into usable information. The foundation for meaningful analysis and insights lies in the data.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*HCF9TJcJpZ11aUIZizlq7A.jpeg" /><figcaption>By Ruqayya Nasir Iro</figcaption></figure><p><strong>Why Does Data Preparation Matter?</strong></p><p>Picture this: You’re a business owner trying to make important decisions based on sales data. But if that data is messy, incomplete, or inaccurate, your decisions could lead to consequences. Data preparation ensures that your data is reliable, consistent, up and ready for analysis.</p><p>Let’s take a look at some real-world scenarios where poor data preparation led to major setbacks:</p><p>1. Misguided Marketing Campaigns: A company launches a marketing campaign based on flawed customer data. As a result, they target the wrong audience, wasting precious resources and missing out on potential sales.</p><p>2. Medical Errors: Inaccurate patient data in the medical field can have fatal results. An example of a nightmare scenario that emphasizes the significance of precise data preparation is when a doctor prescribes the wrong medication due to incomplete medical records.</p><p>3. Financial disasters: Banks and financial institutions rely heavily on data for risk assessment and decision-making. But when data is poorly prepared or contains errors, it can lead to disastrous financial outcomes or identity theft.</p><p>From smart devices and social media to business transactions and healthcare records, we’re generating more data than ever before. But without proper preparation, all that data may be useless. By preparing data, we can fully utilize data-driven technologies such as predictive analytics, machine learning, and artificial intelligence.</p><p>Data preparation may not be the most fun part of the data journey, but it’s undoubtedly one of the most important. By ensuring that our data is clean, consistent, and reliable, we pave the way for meaningful insights and informed decision-making. The next time you are faced with a mountain of data, keep in mind the importance of preparation because it might hold the key to opening up a world of opportunities.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7a4347d1d453" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Making Data Less Boring, More Exciting with Python!]]></title>
            <link>https://medium.com/@rukynas08/making-data-less-boring-more-exciting-with-python-f803e1dc7950?source=rss-f448c518842e------2</link>
            <guid isPermaLink="false">https://medium.com/p/f803e1dc7950</guid>
            <dc:creator><![CDATA[Ruqayya Nasir Iro]]></dc:creator>
            <pubDate>Mon, 26 Feb 2024 14:59:19 GMT</pubDate>
            <atom:updated>2024-02-26T14:59:19.046Z</atom:updated>
            <content:encoded><![CDATA[<p>A world full of data can make it seem like you are trying to find your way through a maze while wearing a blindfold. But worry not, Python is here with its magical abilities for data visualization to rescue the day! Python can turn boring statistics into visually appealing graphics with a simple wave of its code, even for those of us who are not very fond of data.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9tdcZJ2aGPogtkwzcng-4A.jpeg" /><figcaption>Designed by Ruqayya Nasir iro using designer</figcaption></figure><p>Imagine yourself completely submerged in spreadsheets, unable to distinguish between very nice presentations! With tools like Matplotlib and Seaborn at its disposal, Python creates graphs and charts very quickly. Python’s visualizations become more than just pretty pictures; they’re storytelling masterpieces!</p><p>And let’s not forget about the power of interactive visualizations. With Plotly at its disposal, Python can create dynamic charts that let you explore your data. Python produces insightful visualizations that are more than just eye candy! Python helps you find patterns, identify outliers, and make confident decisions by breaking down data into easily absorbed chunks. It is amazing how much fun data analysis can be.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MQo7YWoSXgqiTrA6VOCSLg.jpeg" /><figcaption>Designed by Ruqayya Nasir iro using designer</figcaption></figure><p>In today’s fast-paced world, where attention spans are short, the ability to communicate data effectively is more important than ever. Ultimately, the goal is to uncover the mysteries found in your data and use them to make a meaningful impact on the real world, not just to create attractive charts and graphs.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f803e1dc7950" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Pandas in the Modern World]]></title>
            <link>https://medium.com/@rukynas08/pandas-in-the-modern-world-b1034e277987?source=rss-f448c518842e------2</link>
            <guid isPermaLink="false">https://medium.com/p/b1034e277987</guid>
            <dc:creator><![CDATA[Ruqayya Nasir Iro]]></dc:creator>
            <pubDate>Sun, 11 Feb 2024 14:49:45 GMT</pubDate>
            <atom:updated>2024-02-11T14:49:45.811Z</atom:updated>
            <content:encoded><![CDATA[<p>As our society becomes increasingly data-driven, organizations and individuals increasingly rely on data analysis to gain insights and make decisions. Working with tabular data, however, can be difficult without the right tools. Pandas, an important Python package, makes data analysis and manipulation easier.</p><p>Pandas is not just a tool for scientists and data analysts; it has applications in many other industries as well, revolutionizing decision-making and processes in all areas.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/820/1*C51MnUzR6D6pAqgNFBnM6A.png" /></figure><p>Pandas is used in finance to analyze stock market data, evaluate risk, and create investment strategy prediction models. Pandas is used by banks and other financial institutions for portfolio management, trading algorithm optimization, and fraud detection.</p><p>Pandas are used in healthcare to help with patient data analysis, disease trend identification, and treatment outcome prediction.</p><p>In marketing and e-commerce, Pandas helps businesses understand customer behavior, optimize pricing strategies, and personalize marketing campaigns. Businesses can increase sales and boost customer satisfaction by customizing their offerings to the demands and preferences of their target market through the analysis of customer data.</p><p>Pandas offers a flexible platform for data analysis that makes research in academia easier to conduct across disciplines. Pandas enables researchers to explore and derive meaningful conclusions from their datasets, from studying social trends to analyzing climate data.</p><p>Pandas are useful in daily life as well. Pandas helps people make data-driven decisions in their personal lives, from tracking fitness goals to analyzing household budgets.</p><p>In conclusion, Pandas is more than just a Python library; it is a modern-day innovation and progress accelerator. Its ease of use, versatility, and powerful capabilities make it indispensable across industries and disciplines.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b1034e277987" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Using Numpy for Data Analysis in the Modern Era]]></title>
            <link>https://medium.com/@rukynas08/using-numpy-for-data-analysis-in-the-modern-era-b71edd65af29?source=rss-f448c518842e------2</link>
            <guid isPermaLink="false">https://medium.com/p/b71edd65af29</guid>
            <dc:creator><![CDATA[Ruqayya Nasir Iro]]></dc:creator>
            <pubDate>Sun, 11 Feb 2024 14:11:10 GMT</pubDate>
            <atom:updated>2024-02-11T14:11:10.199Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/952/1*QHEyXoZSSdIEWF3RtkVpAw.png" /></figure><p>Data is everywhere in today’s digital age, from business operations to scientific research. To make informed decisions, it is crucial to analyze this vast amount of data efficiently and accurately. In this context, tools such as Numpy can be very helpful, revolutionizing how we deal with numerical data.</p><p>Numpy, short for Numerical Python, is a fundamental library in Python. Numpy’s ability to handle multidimensional arrays and provide a broad range of mathematical functions for effective manipulation is one of its key features.</p><p>The speed of Numpy is one of its main advantages. Large datasets can be handled with Numpy with great efficiency because of its C implementation, which makes it much faster than pure Python when performing array operations.</p><p>Another important feature of Numpy is its broadcasting capability, which simplifies operations on arrays of different shapes. Numpy makes it easy to perform element-wise operations in place of explicitly writing loops, which improves the readability and conciseness of your code.</p><p>Furthermore, Numpy’s vast mathematical function library meets a variety of needs, ranging from sophisticated linear algebra and statistical analysis to simple arithmetic. Numpy also integrates with other Python libraries like Matplotlib and SciPy, enabling users to visualize data and perform advanced scientific computing tasks with ease.</p><p>In summary, Numpy’s array processing capabilities, performance, and extensive library of mathematical functions make it a crucial tool for data analysis in the modern age. With Numpy, you can handle numerical data with speed and precision for business analytics, scientific research, or machine learning applications.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b71edd65af29" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Ethics in the Age of Data: Lessons from the Equifax Data Breach]]></title>
            <link>https://medium.com/@rukynas08/ethics-in-the-age-of-data-lessons-from-the-equifax-data-breach-623f9010118d?source=rss-f448c518842e------2</link>
            <guid isPermaLink="false">https://medium.com/p/623f9010118d</guid>
            <dc:creator><![CDATA[Ruqayya Nasir Iro]]></dc:creator>
            <pubDate>Thu, 25 Jan 2024 19:48:21 GMT</pubDate>
            <atom:updated>2024-01-25T19:48:21.485Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/774/1*W-yIA0IcriPQy-vhUnDA1A.png" /></figure><p>Data science and machine learning require a strong understanding of data ethics. A common data ethics challenge is data privacy, which refers to protecting personal information. An important aspect of data privacy is the protection of personal information against unauthorized access, use, or disclosure. As data breaches and cyberattacks become more frequent, data privacy is a significant concern today.</p><p>A real-world example of a data privacy breach occurred in 2017 when Equifax, one of the largest credit reporting agencies in the United States, suffered a massive data breach. The breach exposed the private records (including their names, social security numbers, birth dates, and addresses) of 147.9 million Americans, 15.2 million British citizens, and about 19,000 Canadian citizens, making it one of the largest cybercrimes related to identity theft. It was the company’s web application software that had a vulnerability, allowing hackers to access sensitive data. It resulted in identity theft, financial losses, and reputational damage to Equifax.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ajTGoRNnCkOngmmo1JEpZg.png" /></figure><h3>One may ask these questions:</h3><ul><li>How could Equifax have prevented the breach?</li><li>What measures can be taken to ensure that personal data is protected?</li><li>What are the consequences of such a breach?</li></ul><h3>To prevent such breaches, companies can implement measures such as:</h3><ul><li>Keeping software and security protocols up to date.</li><li>Regularly auditing and assessing security vulnerabilities.</li><li>Multi-factor authentication should be implemented.</li><li>Encrypting sensitive data.</li><li>Limiting access to sensitive data.</li></ul><h3>Here are some resources related to the Equifax data breach:</h3><ul><li><a href="https://en.wikipedia.org/wiki/2017_Equifax_data_breach">Equifax data breach — Wikipedia</a></li><li><a href="https://consumer.ftc.gov/consumer-alerts/2019/07/equifax-data-breach-settlement-what-you-should-know">Equifax Data Breach Settlement</a></li><li><a href="https://www.ie.edu/insights/articles/ethical-challenges-in-the-data-world/">Ethical Challenges in the Data World</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=623f9010118d" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>