<?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 Ashish Pahwa on Medium]]></title>
        <description><![CDATA[Stories by Ashish Pahwa on Medium]]></description>
        <link>https://medium.com/@ashishpahwa7?source=rss-f6a19f0f959e------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*75amVgpb9rWuf7M8Grqwww.jpeg</url>
            <title>Stories by Ashish Pahwa on Medium</title>
            <link>https://medium.com/@ashishpahwa7?source=rss-f6a19f0f959e------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 21:32:27 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@ashishpahwa7/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[Empowering Access to Technology: ReUpyog — Your Marketplace for Renewed Computers]]></title>
            <link>https://medium.com/@ashishpahwa7/empowering-access-to-technology-reupyog-your-marketplace-for-renewed-computers-14be56116322?source=rss-f6a19f0f959e------2</link>
            <guid isPermaLink="false">https://medium.com/p/14be56116322</guid>
            <category><![CDATA[non-profit-organization]]></category>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[education]]></category>
            <category><![CDATA[circulareconomy]]></category>
            <category><![CDATA[refurbished-laptops]]></category>
            <dc:creator><![CDATA[Ashish Pahwa]]></dc:creator>
            <pubDate>Sat, 02 Sep 2023 13:10:46 GMT</pubDate>
            <atom:updated>2023-09-02T13:10:46.862Z</atom:updated>
            <content:encoded><![CDATA[<h3>Empowering Access to Technology: ReUpyog — Your Marketplace for Renewed Computers</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ApJqu6XDaaKr1X1Sbrzq6g.png" /></figure><p>In today’s digitally connected world, access to reliable technology is a fundamental need. ReUpyog is not just about selling computers; it’s a marketplace that empowers individuals, educational institutions, nonprofits, and those on a tight budget to access affordable and sustainable technology solutions. In this blog, we’ll delve into how ReUpyog’s marketplace revolutionizes the way we think about tech access, making it inclusive and eco-conscious.</p><p><strong>1. Reupyog: More Than Just Computers</strong></p><p>At Reupyog, we understand that access to technology goes beyond simply buying a computer. It’s about creating a platform that connects people with affordable, reliable devices while championing sustainability. Here’s how Reupyog accomplishes this:</p><p><strong>A Wide Ecosystem:</strong> ReUpyog is your gateway to a diverse ecosystem of renewed laptops, where quality, affordability, and sustainability intersect. Beyond laptops, you’ll find a range of tech solutions to meet your unique needs.</p><p><strong>Inclusivity:</strong> Our marketplace is designed to be inclusive, ensuring that everyone, regardless of their financial situation, has access to technology that can enrich their lives, support their education, or further their missions.</p><p><strong>Eco-Friendly Initiatives:</strong> ReUpyog’s commitment to sustainability extends beyond laptops. We’re dedicated to reducing electronic waste and fostering a culture of responsible consumption in the tech industry.</p><p><strong>2. Serving Students, Nonprofits, and Thrifty Individuals</strong></p><p><strong>Educational Empowerment:</strong> ReUpyog’s marketplace is a lifeline for students. Whether you’re pursuing higher education or engaged in remote learning, our platform connects you with budget-friendly laptops that are perfect for your academic journey.</p><p><strong>Supporting Nonprofits:</strong> Nonprofit organizations can redirect their limited budgets toward their core missions by accessing affordable technology solutions through ReUpyog. Our marketplace aligns with their values of doing more with less.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*V7OtuKbvX6zOZrzcNZ3qEg.png" /></figure><p><strong>Accessible Affordability:</strong> For individuals and families looking for economical tech solutions, Reupyog provides a range of options suitable for various needs, from online learning to remote work and personal use.</p><p><strong>3. Join the ReUpyog Community</strong></p><p>ReUpyog is more than a marketplace; it’s a community that believes in the power of accessible technology and sustainability. Join us to:</p><p><strong>Promote Sustainable Practices:</strong> By choosing renewed laptops and tech accessories from ReUpyog, you play a crucial role in reducing electronic waste and promoting a greener, more sustainable world.</p><p><strong>Unlock Opportunities:</strong> Access affordable and reliable technology that can empower you, your organization, or your family to thrive in the digital age.</p><p>Join our community today and experience the difference of ReUpyog, where technology is more than a commodity; it’s a vehicle for empowerment and positive change. Empower yourself and your community with Reupyog’s marketplace for renewed laptops and more.</p><p>Visit: <a href="https://reupyog.in/">https://reupyog.in/</a><br>Or <a href="https://reupyog.in/shop"><em>https://reupyog.in/shop</em></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=14be56116322" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Revolutionizing Consumption Habits: Circular Economy Vision and Youth Leadership in India]]></title>
            <link>https://medium.com/@ashishpahwa7/revolutionizing-consumption-habits-circular-economy-vision-and-youth-leadership-in-india-8cae5b937b13?source=rss-f6a19f0f959e------2</link>
            <guid isPermaLink="false">https://medium.com/p/8cae5b937b13</guid>
            <category><![CDATA[environment]]></category>
            <category><![CDATA[circulareconomy]]></category>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[sustainability]]></category>
            <dc:creator><![CDATA[Ashish Pahwa]]></dc:creator>
            <pubDate>Thu, 24 Aug 2023 20:06:43 GMT</pubDate>
            <atom:updated>2023-08-24T20:06:43.038Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LAPs2dktyM9dK62J7U6U6Q.png" /><figcaption>ReUpyog was selected as one of the 10 startups to showcase India’s diverse innovative power to German business delegation, which include Dr. <a href="https://www.linkedin.com/in/ACoAAEEKqfgBzO-ud8D4PTvoheM9JQ4zUJEJpoc">Robert Habeck</a>, the Vice Chancellor and Federal Minister for Economic Affairs and Climate Action of Germany, and esteemed companies such as Siemens, BASF, MERCK, Infineon, SAP and German media.</figcaption></figure><p>In a world grappling with environmental challenges, the passionate efforts of India’s youth, exemplified by strong solution based climate movements like the <a href="https://iycn.in/">Indian Youth Climate Network</a>, are shaping a formidable force for change. As India navigates its unique environmental challenges, circular economy business platforms like <a href="https://reupyog.in/">ReUpyo</a>g have emerged as dynamic solutions that not only support responsible consumption but also generate job opportunities, amplifying the impact of the youth-led movement.</p><p><strong>Expanding Horizons Beyond Energy with ReUpyog:</strong></p><p>As the call for a just transition reverberates, it becomes clear that this transformation must extend beyond the realm of energy. This is where <a href="https://reupyog.in/">ReUpyog’s Circular Economy Marketplace</a> comes into play, championing a holistic approach to sustainability. The ethos of ReUpyog urges us to reconsider our approach to products — encouraging us to prioritize repair-ability, refurbishment, and repurposing. In doing so, ReUpyog echoes India’s culture of resourcefulness and emphasizes a transition that benefits both the environment and society.</p><p><strong>Empowerment through Repair and ReUpyog:</strong></p><p>In the midst of today’s fast-paced consumer culture, the notion of the “right to repair” gains immense significance. It’s a concept that ReUpyog wholeheartedly embraces. By offering renewed and refurbished products, ReUpyog extends to individuals the power to actively contribute to sustainability. Through ReUpyog, consumers are empowered to counteract the throwaway mindset and instead opt for choices that align with the principles of longevity and resource conservation.</p><p><strong>Circular Economy: Beyond Sustainability</strong></p><p>While circular economy marketplaces are often associated with selling reused products, their essence reaches far beyond that. These platforms embody a multifaceted approach to sustainability that extends beyond resource conservation. They also act as potent hubs for job creation. These platforms generate employment opportunities through various stages of the product life cycle — repair, refurbishment, repurposing, and even distribution. This interconnected network not only champions sustainability but also contributes to economic growth by engaging skilled artisans, technicians, and entrepreneurs.</p><p><strong>Navigating Waste Management with ReUpyog’s Insight:</strong></p><p>It’s important to recognize that while circular economy solutions are often hailed for their sustainable qualities, they also play a pivotal role in efficient waste management. By diverting items away from landfills and incinerators, ReUpyog not only promotes sustainability but also eases the burden on waste disposal systems. This twofold impact highlights how circular economy practices, as facilitated by ReUpyog, seamlessly merge with effective waste management strategies.</p><p><strong>A Wholesome Perspective with ReUpyog’s Vision:</strong></p><p>ReUpyog’s circular economy marketplace epitomizes a comprehensive approach to sustainability — one that transcends transactions and embraces a broader ethos. As the youth actively engage with ReUpyog, they step into the role of Agents of Change. ReUpyog doesn’t merely offer products; it presents a movement — a movement that advances sustainable living, conscious decisions, and a harmonious coexistence with the environment.</p><p>In India’s quest for sustainability, ReUpyog’s circular economy marketplace emerges as a canvas of possibilities. Driven by the vigor of the youth and fortified by the ideals of a just inclusive transition,</p><p><strong>Navigating Challenges in the Circular Economy Vision:</strong></p><p>However, the road to a holistic circular economy vision is not without challenges. Creating a value chain where none existed before, standardizing practices, and decentralizing operations are intricate tasks. Ensuring quality control and managing demand and supply in sync is a complex puzzle. At ReUpyog, we are working against these challenges, redefining systems and approaches, and building an ecosystem that thrives on collaboration.</p><p><strong>Changing Behaviors in the Informal Sector:</strong></p><p>One significant challenge lies in changing the behavior of the informal sector. To encourage sustainable practices here, substantial incentives are necessary. Recognizing the informal sector’s contribution and empowering it to embrace circular practices requires innovative policies, economic incentives, and education. ReUpyog’s holistic approach includes the informal sector as a vital player in this transformative journey, offering them a platform to grow sustainably.</p><p>ReUpyog invites all to partake in a journey that propels us toward a more resilient and sustainable India — one that embraces innovation, conservation, and the promise of a greener tomorrow.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8cae5b937b13" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[In Depth Statistical Analysis of  VGG Face Network’s Performance on Custom Data set]]></title>
            <link>https://medium.com/analytics-vidhya/in-depth-statistical-analysis-of-vgg-face-networks-performance-on-custom-data-set-ef4b72bc15b9?source=rss-f6a19f0f959e------2</link>
            <guid isPermaLink="false">https://medium.com/p/ef4b72bc15b9</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[statistics]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[face-recognition]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Ashish Pahwa]]></dc:creator>
            <pubDate>Sun, 20 Oct 2019 18:06:01 GMT</pubDate>
            <atom:updated>2019-11-18T06:35:04.748Z</atom:updated>
            <content:encoded><![CDATA[<h3>In Depth Statistical Analysis of VGG Face Network’s Performance on Custom Data set</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*L8hcB-HcYuFHagCn" /><figcaption>Photo by <a href="https://unsplash.com/@bmowinkel?utm_source=medium&amp;utm_medium=referral">Brandon Mowinkel</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>VGG-Face is a very deep CNN architecture which learned on million of celebrity images. It is being used as feature extractor to extract the activation vector of the fully connected layer. I have been going through a lot of blogs that explain about face verification and identification on your custom data set. I have mentioned few of them. Hope you’ll find them useful too.</p><ul><li><a href="https://gist.github.com/EncodeTS/6bbe8cb8bebad7a672f0d872561782d9">VGG-Face Model for Keras</a></li><li><a href="https://machinelearningmastery.com/how-to-perform-face-recognition-with-vggface2-convolutional-neural-network-in-keras/">How to perform Face Recognition</a></li><li><a href="https://www.dlology.com/blog/live-face-identification-with-pre-trained-vggface2-model/">Live Face Detection</a></li></ul><p>In this post I’ll talk about how to come up with a similarity index threshold (particularly cosine similarity score) that is optimal for your own data. Depending upon the use case you might have to revise it (or let it revise itself automatically!) as you start collecting more data.</p><p>Let me begin by setting up the stage-</p><p>To begin with I collected around 200 faces of 5 people for experimental purpose. Then I prepared pandas data frame to perform pairwise comparison of images within each groups well as across groups.</p><p>Let’s say there are k categories or groups. Each group consist of multiple face images of a single person.</p><p>n₁ = number of images of in group 1, n₂ = number of images of in group 2 and so on.</p><p>Then number of unique pairwise comparisons within groups can be computed using:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/116/1*TyvVNcD5BKgh32sQpTXyIw.png" /></figure><p>Since there are are k groups there are kC₂ unique pairs of groups. Total number of inter group comparisons will then be,</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/210/1*pyKClQZU960KL8Pa5sxneg.png" /></figure><p>To sum up :</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/329/1*f_UlkWzIBthP5Mu7cWmxxA.png" /></figure><p>To make life a bit simpler, I sampled (without replacement) 10% of pairs from each group and 10% from each set of inter group pairs. So I was left with around 2000 pairs of images.</p><p>After performing 2000 pairwise comparisons and appending the results to pandas data frame, the data set looked as shown in the figure:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/732/1*XzMnN-pfaOriM1joRrsE2g.png" /></figure><p><strong>Data Description:</strong></p><p>Img1: Path of image1</p><p>Img2 : Path of Image 2</p><p>is_match : Truth value of whether the two faces belong to the same person</p><p>epsilon: cosine similarity score</p><p>true_class: name of the person</p><p>In the data generation process, I ensured there are no missing values inserted in the data set.</p><p><strong>Exploratory Data Analysis</strong></p><p>A. Comparative Frequency Distribution</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/562/0*xJr3wyLMpfakKh9j" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/587/0*ngMbJTUqN3YshWWx" /></figure><p><strong>Key Observations:</strong></p><ul><li>It “seems” like the two means are different from each other.</li><li>For negative class, it is easy to see that the distribution is approximately normal.</li><li>If images are identical, the mean tends to shift towards 0.</li><li>There are some outliers in class labeled ‘True’. (We don’t want high epsilon values to be labelled True)</li></ul><p>Let’s explore how the system behaves for different groups. (When the images are identical)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/649/0*f4ESC36daBtNWY2J" /></figure><p><strong>Key Observations</strong>:</p><ul><li>For C and D means are nearly equal.</li><li>For all faces that have been identified as identical cosine score are less than 0.4</li><li>Question — Does that means 0.4 should be a decent optimal value for a threshold parameter ? Let’s explore in more detail.</li></ul><p>We still need to understand whether the observed differences between the sample means simply due to sampling variation or due to real differences in the population means?</p><p>Consequently we’d also like to answer the following questions before jumping to model building step.</p><ul><li>Is the difference between epsilon values of positive and negative classes statistically significant ?</li><li>Is the system biased towards any particular class?</li></ul><p><strong>Analysis of Variance</strong></p><p>One Way analysis of variance (ANOVA) is used to compare several means. This method is often used in scientific or medical experiments when treatments, processes, materials or products are being compared.</p><p>Comparing Group Means with Median:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/629/1*-VC4FStTznLXB6ub260JEA.png" /></figure><p>It’s clear that, we have a symmetric distributions across groups. So first assumption is satisfied.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/289/1*5SKE7RsBD5HWlP-k5MRkUw.png" /></figure><p>It also turned out that group standard deviations are quite similar. These results, together with the box plots given above, indicate that the distribution of each group is symmetric. Therefore, we can proceed with the analysis of variance. Due to lack of data points I have intentionally dropped rows containing F class.</p><p><strong>ANOVA Table</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/692/1*XahL6nybKuwNb9Bcerb8NQ.png" /></figure><p>From our results we can say that there is strong evidence that the mean epsilon value varies with group. There exists at least one group whose mean is unequal.</p><p><strong>Tukey HSD Test:</strong></p><p>Tukey HSD test shows that the differences between classes A,B and C are significant at 95% confidence level. However for all the other pair of groups there is no significant difference between the mean values of epsilon score. That implies VGG network might not be able to identify faces in these groups correctly in some cases. Probably due to poor image quality or lightning conditions? That requires further investigation. We wouldn’t expect CNN to be biased towards certain group of people.</p><p><strong>Train and Test Dataset</strong></p><p>The generated data set is divided into Training and Testing data set, with 70:30 proportion. The distribution of positive and negative Class is verified in both the data sets, and ensured it’s within tolerance limit.</p><p><strong>Modelling — Logistic Model</strong></p><p>Following is the output of my logistics model:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/703/1*kWUAIb6x4aHljespKCqvXw.png" /></figure><p>Number of Fisher Scoring iterations = 6, AIC : 1075</p><p>It is clearly evident that epsilon is highly significant in determining whether the two faces are identical or not.</p><p>The plot below depicts that low epsilon scores contribute towards high probability of match.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/588/0*heNwt5_f_l_y5I61" /></figure><p><strong>ROC Curve</strong></p><p>ROC curve is a performance measurement for classification problem at various thresholds settings. Area under the curve obtained by this model at threshold of 0.2 is 0.8349. Higher the AUC, the better the model is at distinguishing between identical and dissimilar faces.</p><p>So how good is it ? How do you determine that threshold is good for your model? Let’s explore it carefully.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/296/0*JwY6KX9chapqoprV" /></figure><p><strong>Confusion Matrix</strong></p><p><strong>(</strong>At threshold = 0.3<strong>)</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/495/1*dRLs0i9NWXNIpRx8_8qADg.png" /></figure><p>Overall Accuracy: (338+73)/(338+73+74+38) = 78%</p><p>Now here things get a little trickier.</p><p>Should we accept this model and deploy it on our production system?</p><p>At this point you should first carefully analyze and understand the purpose of your system and add some business context to it. What is more important to you — To label faces as identical given they are actually identical with high accuracy? What is your tolerance to labeling face a match if it’s not actually match?</p><p>For example the above model labels a pair of faces as False if they are actually dissimilar 91% of the times. Which may be good for a certain use case.</p><p>Equipped with the above statistical understanding, now you can fine tune your threshold depending on the requirement of the hour. Such statistical modelling techniques allow you to establish performance standards, customize your face detection model for various groups if required or simply just monitor the performance in real world scenario.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ef4b72bc15b9" width="1" height="1" alt=""><hr><p><a href="https://medium.com/analytics-vidhya/in-depth-statistical-analysis-of-vgg-face-networks-performance-on-custom-data-set-ef4b72bc15b9">In Depth Statistical Analysis of  VGG Face Network’s Performance on Custom Data set</a> was originally published in <a href="https://medium.com/analytics-vidhya">Analytics Vidhya</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[First Case Study in Descriptive Analytics]]></title>
            <link>https://medium.com/@ashishpahwa7/first-case-study-in-descriptive-analytics-a744140c39a4?source=rss-f6a19f0f959e------2</link>
            <guid isPermaLink="false">https://medium.com/p/a744140c39a4</guid>
            <category><![CDATA[statistics]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[descriptive-analytics]]></category>
            <dc:creator><![CDATA[Ashish Pahwa]]></dc:creator>
            <pubDate>Fri, 10 May 2019 17:46:52 GMT</pubDate>
            <atom:updated>2019-05-11T02:23:01.645Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/256/1*obS521FYknmCZGFEUtrL9g.png" /></figure><p>This is my fourth blog in the series. As promised, I am presenting the first case study based on some of the concepts discussed in previous posts.</p><p>Market research team at Aqua Analytics Pvt. Ltd is assigned a task to identify profile of a typical customer for a Digital fitness band that is offered by Titanic Corp. The market research team decides to investigate whether there are differences across the usage patterns and product lines with respect to customer characteristics. The team collected data of customers who purchased the band from distributor during the prior six months.</p><p>The team identified following customer variables to study:</p><ul><li>Gender</li><li>Age (In years)</li><li>Education(In years)</li><li>Relationship Status(Single or Partnered)</li><li>Annual Household income</li><li>Average number of times customer tracks activity each week</li><li>Number of miles customer expect to walk each week</li><li>Self-rated fitness on a scale 1–5 where 1 is poor shape and 5 is excellent</li></ul><p>Models of the product purchased — IQ75, MZ65, DX87</p><p>Perform descriptive analytics to create a customer profile for each model’s product line.</p><p>Here’s how the first few rows of data set looks like.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/507/0*75zPeQ11VMRlgEWk" /><figcaption>Figure 1</figcaption></figure><p>Let’s also explore how the profiling characteristics differ with respect to gender.</p><p><strong>Male:</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/654/0*LMurncWU1UAP8E4o" /><figcaption>Figure 2</figcaption></figure><p><strong>Female:</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/654/0*P9qnUI0WteBGLsMj" /><figcaption>Figure 3</figcaption></figure><p>33 males are using DX87 but only 7 females are using this model. Why is that? May be research and development teams would like to study it in more depth after getting this insight.</p><p>Male users have higher median income than female customers. Both males and females seem to have around same usage frequency, similar education levels and they both seem to fall in same age group.</p><p>Let’s perform some visual analytics using pattern recognition tools.</p><p>The histogram represents the age distribution of all customers. It is a right skewed distribution having more people falling the age bucket of 20–25.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*XwcmH-czviA0gu5C" /><figcaption>Figure 4</figcaption></figure><p>Standard Deviation : 6.9</p><p>Mean : 28.7</p><p>Let’s look at the boxplot:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*n2DcEjvca8LBW2Gr" /><figcaption>Figure 5</figcaption></figure><p>It is clearly evident that most values lie towards the right of median.</p><p>Let’s ask another question — Are there difference in ages of men and women?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*xh1O4ERjrk2Qm6Q3" /><figcaption>Figure 6</figcaption></figure><p>I think the answer is No. Box plots suggest they are more or less same. We can also confirm this from crosstabs.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/162/0*p8nNPkT_SzxdErV-" /><figcaption>Figure 7</figcaption></figure><p>Notice that DX87 model is dominated by males! That’s a great insight.</p><p>Let’s also try to explore if there are any differences in choice of model with respect to age</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*AW0lPyDZvRu9Ej_D" /><figcaption>Figure 8</figcaption></figure><p>The comparison of box plots suggest that there definitely exists some difference in the age group of customers who are using DX87 model. These customers seems to be younger than the the ones who are using MZ65 and IQ75.</p><p>We can also see how the preference of model varies depending on marital status.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/184/0*gyZqBSvMFuUisbNz" /><figcaption>Figure 9</figcaption></figure><p>It is observed that more partnered people are using each design.</p><p>Let’s have a look at some usage statistics:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/706/0*tx-nZdMac7kGnCgh" /><figcaption>Figure 10</figcaption></figure><p>Most of the customers use it 3–4 times a week.No female consumer has ever used IQ75 and MZ65 more than 5 time a week.</p><p>PS: This case study is inspired from a case study in <a href="https://www.amazon.in/Business-Statistics-Viswanathan-Stephan-Kathryn-ebook/dp/B074PMSB9C/ref=sr_1_1?qid=1557508546&amp;refinements=p_27%3AP.+K.+Viswanathan%2C+David+Stephan+%2C+Kathryn+A+Szabat+David+M.+Levine&amp;s=digital-text&amp;sr=1-1&amp;text=P.+K.+Viswanathan%2C+David+Stephan+%2C+Kathryn+A+Szabat+David+M.+Levine">Business Statistics : A First Course by Dr. P. K. Viswanathan, David Stephan , Kathryn A Szabat David M. Levine</a></p><p>Code:</p><blockquote>data = read.csv(“data.csv”, header=TRUE)</blockquote><blockquote>attach(data)</blockquote><blockquote>View(data)</blockquote><blockquote>summary(data)</blockquote><blockquote>by(data, INDICES = Product, FUN=summary)</blockquote><blockquote>by(data, INDICES = Gender, FUN=summary)</blockquote><blockquote>hist(Age, col=”red”) # Age distribution show pattern of not a bell curve</blockquote><blockquote>boxplot(Age, horizontal = TRUE, col=”blue”, main=”Box Plot of Age”)</blockquote><blockquote>boxplot(Age~Gender, horizontal=TRUE, col=c(“red”,”blue”),main=”comparitive boxplot”)</blockquote><blockquote>boxplot(Age~Product, horizontal=TRUE)</blockquote><blockquote>table(Product, Gender)</blockquote><blockquote>sd(Age)</blockquote><blockquote>mean(Age)</blockquote><blockquote>table(Product, MaritalStatus)</blockquote><blockquote>library(rpivotTable) # simply outperforms excel in terms of ease of operations</blockquote><blockquote>rpivotTable(data)</blockquote><p>This is just a tiny crystal at the tip of an ice berg, you’ll be amazed to see what all you can achieve through some limited number of tools and statistically techniques. Some of them are really powerful and goes beyond just explaining the sample statistics. I’ll cover them in this series soon :)</p><p>Happy Learning</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a744140c39a4" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Measures of Dispersion]]></title>
            <link>https://medium.com/@ashishpahwa7/measures-of-dispersion-d34655364a71?source=rss-f6a19f0f959e------2</link>
            <guid isPermaLink="false">https://medium.com/p/d34655364a71</guid>
            <category><![CDATA[business-intelligence]]></category>
            <category><![CDATA[business-analytics]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[statistics]]></category>
            <dc:creator><![CDATA[Ashish Pahwa]]></dc:creator>
            <pubDate>Sun, 05 May 2019 19:25:39 GMT</pubDate>
            <atom:updated>2019-05-05T19:49:09.964Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>Measures of Dispersion</strong></p><p>This is my third blog in the series of Data Science blogs with applications in business intelligence.</p><p>Mean, median, mode are all summary statistics. In statistics we say, they all define population characteristics. For example suppose I have to measure average income of a particular market segment. But I cannot visit each and every person and ask what is their income. So I cannot compute population mean. However it is feasible to select a comparatively small sample of population and compute its arithmetic mean. This Average is called sample mean. So in a way sample mean income estimates the population income. Let’s say value of sample mean turn out to be 4 lakhs</p><p>But it has a major drawback. It makes sense to ask- Is the income always 4 lakhs?</p><p>Explanations to such questions are provided by Measures of dispersion (also called measures of variation). You need to combine the summary measures with measures of dispersion to give meaning to the data.</p><p><strong>Range:</strong></p><p>Range is the simplest of all measures of dispersion. It depends on only two values in the dataset, i,e. maximum and minimum</p><p>Range = maximum — minimum</p><p><strong>Range is a measure of volatility</strong>. If you visit websites like <a href="https://www.moneycontrol.com/stocks/marketstats/bsehigh/index.php?classic=true">moneycontrol.com</a>, you’ll notice they alway highlight 52 week high/ 52 week low values. They represent the highest and lowest price at which a stock has traded during the previous year and is extremely important in determining stock’s current value and predicting future movement.</p><p><strong>Drawback of Range:</strong></p><p>If any of the value ( maximum or minimum ) turns out to be an outlier, the range will collapse.</p><p><strong>IQR Range</strong></p><p>It is also called mid range. It is range of the middle 50% of observation. First and last 25% of the observations are not considered. It is computed using a statistical formula:</p><p>IQR = Q3 — Q1</p><p>where,</p><p>Q3 : Third quartile or 75th percentile</p><p>Q1 : First quartile or 25th percentile</p><p>IQR has tremendous resistant to outliers.</p><p><strong>Standard Deviation</strong></p><p>Last but not the least we have Standard deviation which forms the cornerstone of Inferential Statistics. To define standard deviation you need to define another term called variance. In simple terms, standard deviation is the square root of variance.</p><p>Example:</p><p>The following data represent the percentage return on investment for 10 mutual funds per annum (B3 to B12)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/435/0*4qOQfWAdwMScvVu8" /></figure><p>Standard deviation is the measure of risk. When you talk about mutual funds or other stocks, your aim at minimizing the risk and at the same time you have to take some risk. So what is the best portfolio?</p><p>High risk, high return,</p><p>Medium risk, medium return</p><p>Low risk, Low return</p><p>In all the stocks, you must have combination of all these three baskets. So what you should do is pick up some stocks that are risky, pick up some that are not that risky and pick up some that have low risk. So the overall risk is minimized.</p><p>So the the standard deviation can help you choose an optimal portfolio of stocks if you have to invest in them.</p><p>Caution : Standard deviation also depends on the original scale. So let’s say you are studying currency, and you are operating in dollars, then standard deviation will also be in dollars. If you have to compare different countries, then you should use coefficient of variation, which is the relative standard deviation.</p><p><strong>Coefficient of Variation</strong></p><p>Coefficient of Variation (CV) is defined as the ratio of Standard Deviation to Mean.</p><p><strong>Another Application</strong></p><p>Let’s say one sales person, X gets 50 units average sales, with standard deviation of 5 and there is another sales person, Y who made an average sales of 75 units with standard deviation 25.</p><p>Then CV of X = 0.10 or 10% and CV of Y = 0.33 or 33%</p><p>So who is better?</p><p><strong>Moral of the story </strong>: Don’t get carried away by absolute number, Look at the scatter. Even though Y has achieved a higher average, his performance is not consistent. But that does not mean you don&#39;t need X, you do want higher sales, but you should advice him to be consistent.</p><p>In the next post, I will discuss about the objectives of Descriptive Statistics using a case study.</p><p>Get ready to get your hands dirty on some code!.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d34655364a71" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Setting up The Stage for Analytics]]></title>
            <link>https://medium.com/@ashishpahwa7/setting-up-the-stage-for-analytics-971b327ef61b?source=rss-f6a19f0f959e------2</link>
            <guid isPermaLink="false">https://medium.com/p/971b327ef61b</guid>
            <category><![CDATA[business-analytics]]></category>
            <category><![CDATA[business-intelligence]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Ashish Pahwa]]></dc:creator>
            <pubDate>Wed, 01 May 2019 14:14:00 GMT</pubDate>
            <atom:updated>2019-05-01T14:16:41.777Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*bX2Uv4dbh49q0CJTSYjAOg.jpeg" /></figure><p>This is my second post in the series of data science blogs. Before moving into technical aspects, I thought of sharing some background knowledge that may help you visualize problems in a way different than you already do, especially if you are a beginner coming from a technical background.</p><p>Many beginners tend to underestimate the importance of formalizing the problem that needs to be solved and lack the understanding of ultimate business objectives. It may be because during our learning journey, most of the time we had problems well framed and data readily available. Another reason is that setting up a predictive modelling task requires broad set of skills — knowledge of business domain, databases or data infrastructure, and predictive modelling algorithms.The Cross-Industry Standard Process Model for Data Mining (CRISP-DM) describes the data-mining process in six steps. It is documented in a 80 page <a href="https://drive.google.com/open?id=1pzpWGmKjp4sBuAJoVXib1gZPiekGk3xH">PDF </a>file.</p><p>On a high level, the 6 steps are defined as follows:</p><ol><li>Business Understanding</li><li>Data understanding</li><li>Data Preparation</li><li>Modelling</li><li>Evaluation</li><li>Deployment</li></ol><p>Sometimes analysts have business understanding but most often managers and directors have far better perspective on how models affect organization.</p><p><strong>The Three Leg Stool</strong></p><p>You can think of collaboration needed for a successful analytics project as a three legged stool. The three legs are -</p><ol><li>Domain expert</li><li>Database expert</li><li>Predictive Modelling expert</li></ol><p>If anyone of them do not perform their job up-to a defined standard, then analytics project is likely to be failed. For instance database and predictive modelling experts can build highly accurate models but they may not address the actual need of organization.</p><p>It is important to realize that not all models are built to “predict” an outcome. Many a times they are built for the sole purpose of understanding the behavior of a human or a process.</p><p><strong>Understanding Data</strong></p><p>I have used the term data a lot many times now. Let’s see how it actually looks like and how it is arranged. After setting up the stage, understanding your data is the next step which I will discuss in detail in upcoming posts.</p><p>We arrange data in a two-dimensional structure called matrix. Column names are called attributes or independent variables that may be of interest. For example in customer analytics one attributes may be customer id, age, location and so forth.</p><p>Every row contains value, i.e measure of each corresponding attribute. It is worth giving a thought that modeling algorithms assume that each observation is independent of each other. If in a given dataset two records are customers who are brother and sister and they frequently purchase same items together, the algorithm treats them no different than any two different people with similar behavior pattern.</p><p><strong>Target Variable</strong></p><p>For any predictive modelling task it is crucial to identify one or more target variables because it is the only information that predictive algorithm has to uncover. Remember that algorithms do not have common sense and they cannot bring context to the problem in a way humans can. Therefore target variables must be quantified in a way that they can explain as much business objective as possible.</p><p>There are more aspects to understanding data and setting up the problem right than what I have covered in this blog but I think this background should be sufficient enough to head start. I will continue this topic in subsequent posts after covering few technical concepts so you can relate to it even better.</p><p>Keep reading, keep learning.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=971b327ef61b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Introduction To Business Analytics]]></title>
            <link>https://medium.com/@ashishpahwa7/introduction-to-business-analytics-8841510fdd53?source=rss-f6a19f0f959e------2</link>
            <guid isPermaLink="false">https://medium.com/p/8841510fdd53</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[statistical-learning]]></category>
            <category><![CDATA[business-intelligence]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[business-analytics]]></category>
            <dc:creator><![CDATA[Ashish Pahwa]]></dc:creator>
            <pubDate>Tue, 30 Apr 2019 15:39:17 GMT</pubDate>
            <atom:updated>2019-05-01T13:53:47.885Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/800/1*xq8j-ocuWz4NHVzteDLREw.jpeg" /></figure><p>This is my first blog in the series of upcoming Data Science blogs. In this series I will try to cover some of the important concepts in Statistics for decision making, predictive modelling, advanced statistics, data mining, machine learning and time series forecasting supported by case studies and examples. I will also try to cover few topics in marketing as well.</p><p>Most of us know data vs information. Data is a raw material and information is finished product. People talk about insights, but you don’t get insights without information.</p><p>If you look at trial balance in a company, its data, not information. But If I construct a balance sheet using the trial balance then it becomes partial information as people read the balance sheet and get a sense of financial health of a company. But if I analyze the balance sheet and identify some important financial ratios like current ratio,capital structure ratio, profitability ratio, revenue per employee and how these things move over a period of time, then I am getting insights out of the balance sheet. So insights are the fine aspects of the information that helps in decision making.</p><p>Nowadays there is plenty of data and managers are confused about how to utilize that data to improve the profits of organization, they are looking for methods to classify the data so it can make some sense. So to help the management how do we take the right decision or make meaningful inferences from data. Data by itself does not help, but from data if you predict market churn, i.e when will a particular consumer move to another service provider, that becomes very crucial.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/221/0*kJxiNdZP8IQ88cMI" /></figure><p><strong>Japanese have popularized 7 tools of quality:</strong></p><p>1. Check-sheet, where you collect the data in a specific format</p><p>2. Frequency distribution and histogram</p><p>3. Stratification, wherein you put similar objects in different clusters so that homogeneity is seen within cluster and heterogeneity in between clusters.</p><p>4. Scatter diagram which analyses how x behave in relation to y or vice versa</p><p>5. <a href="https://en.wikipedia.org/wiki/Pareto_chart">Pareto diagram</a></p><p>6. <a href="https://en.wikipedia.org/wiki/Ishikawa_diagram">Ishikawa diagram</a></p><p>7. <a href="https://en.wikipedia.org/wiki/Control_chart">Control charts</a>, also known as process behavior chats</p><p>The first tool that is very important after the checksheet is histogram. To get the histogram you need frequency distribution.</p><p><strong>What is frequency distribution?</strong></p><p>Frequency distribution is a table which contains data arranged into classes with the number of objects that fall into each class. The number of observations that fall into each class is called frequency. But the frequency table itself will not give you any pattern, so you need a histogram.</p><p>You can think of histogram as a photograph of frequency distribution, It gives a pattern — whether the distribution is a bell curve, left skewed, right skewed, whether distribution has peaks, so on and so forth.</p><p>In statistics people always talk about the bell curve.</p><p>The bell curve is extremely useful for establishing standards in performance measures(Will discuss about it in future posts) . Therefore bell curve is becoming a prerequisite for many applications.</p><p><strong>Uses of Histogram :</strong></p><p>1. In materials management for assessing materials strength.</p><p>2. We use histogram for estimating process capability. In context of manufacturing, it means is my process capable of meeting customer specification.</p><p>3. If it is not bell curve, it tells you corrective actions must be taken, because it indicates necessity of corrective actions, then after the corrective action is the system or process stable? Histogram will tell you.</p><p>4. Estimating machine capability in terms of speed and performance</p><p>5. We can compare different vendors, products, regions and performance across different locations using histograms.</p><p>To summarize histogram is a pattern recognition tool that tells you the shape of distribution and associated properties which help decision makers take remedial actions.</p><p>Let’s now move to statistical summary measures called statistical averages.</p><p><strong>Statistical Averages</strong></p><p>When I say average, lot of people usually imply arithmetic mean, but that’s not correct.</p><p>There are many statistical averages — Mean, median, mode, harmonic mean, geometric mean etc.</p><p>We use a term here “Measure of Central Tendency”.</p><p>What is central about it? What is tendency ?</p><p>Whenever you take measurement of same characteristics a number of times, you form a dataset.Then we will ask a fundamental question — Is it possible for me to pick up on average out of these measurements in such a manner that remaining items in the data have tendency to be around this value. Such a measure is called measure of central tendency.</p><p>Most often used measures of central tendency are</p><p>1. Arithmetic Mean</p><p>2. Median</p><p>3. Mode</p><p>Important thing to note is that all the three measures must be understood in the context of application. Context is extremely important.</p><p>I am assuming readers of this blog have basic idea of what they are and how they are computed.</p><p><strong>1. Arithmetic Mean:</strong></p><p>Arithmetic Mean is affected by extreme values or fluctuations in sampling. It is not the best average to use when dataset contains extreme values. (Very high or very low values)</p><p>For example let us consider CTC of 11 students ( in lakhs per annum)</p><p>3, 4, 6, 4, 7, 5, 4, 5, 6, 18, 5</p><p>Average = 6</p><p>But if I remove the student who grabbed 18 lpa from the dataset, average is reduced to 4</p><p><strong>2. Median:</strong></p><p>It is the middle value when dataset is arranged in ascending ( or descending ) order of magnitude.</p><p>So where do we use median?</p><p>In market research,consumer preferences study, we always have a consumer ranking ( ordinal ranking). We use median as a reference point in the distribution pattern.</p><p><strong>3. Mode</strong></p><p>Mode is defined as that value that occurs most often. Mode also has resistant to outliers next to median. In Fact both mode and median are capable of fighting outliers better than mean.</p><p>When do we use mode?</p><p>For example, say you go to buy a pair of shoes at Nike, Bata or any shoe company’s outlet. You will notice that most frequently occurring size is always available in the inventory.</p><p>Caution : In few problems in real life, there will be more than one mode such as bimodal and multi modal values. In these cases mode cannot be uniquely determined.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8841510fdd53" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Applications of Data Science in Business Intelligence]]></title>
            <link>https://medium.com/series/applications-of-data-science-in-business-intelligence-2d570d41e50f?source=rss-f6a19f0f959e------2</link>
            <guid isPermaLink="false">https://medium.com/p/2d570d41e50f</guid>
            <dc:creator><![CDATA[Ashish Pahwa]]></dc:creator>
            <pubDate>Tue, 30 Apr 2019 15:33:19 GMT</pubDate>
            <atom:updated>2019-04-30T15:33:19.163Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>Introduction to Business Analytics</strong></p><p>This is my first blog in the series of upcoming Data Science blogs. In this series I will try to cover some of the important concepts in Statistics for decision making, predictive modelling, advanced statistics, data mining, machine learning and time series forecasting and may be few topics in marketing as well.</p><p>Most of us know data vs information. Data is a raw material and information is finished product. People talk about insights, but you don’t get insights without information.</p><p>If you look at trial balance in a company, its data, not information. But If I construct a balance sheet using the trial balance then it becomes partial information as people read the balance sheet and get a sense of financial health of a company. But if I analyze the balance sheet and identify some important financial ratios like current ratio,capital structure ratio, profitability ratio, capital per employee and how these things move over a period of time, then I am getting insights out of the balance sheet. So insights are the fine aspects of the information that helps in decision making.</p><p>Nowadays there is plenty of data and managers are confused about how to utilize that data to improve the profits of organization, they are looking for methods to classify the data so it can make some sense. So to help the management how do we take the right decision or make meaningful inferences from data. Data by itself does not help, but from data if you predict market churn, i.e when will a particular consumer move to another service provider, that becomes very crucial.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/276/0*6mHUg8oYywKKcGN6" /></figure><p><strong>Japanese have popularized 7 tools of quality:</strong></p><p>1. Check-sheet , where you collect the data in a specific format</p><p>2. Frequency distribution and histogram</p><p>3. Stratification, wherein you put similar objects in different clusters so that homogeneity is seen within cluster and heterogeneity in between clusters.</p><p>4. Scatter diagram which analyses how x behave in relation to y or vice versa</p><p>5. <a href="https://en.wikipedia.org/wiki/Pareto_chart">Pareto diagram</a></p><p>6. <a href="https://en.wikipedia.org/wiki/Ishikawa_diagram">Ishikawa diagram</a></p><p>7. <a href="https://en.wikipedia.org/wiki/Control_chart">Control charts</a>, also known as process behavior chats</p><p>The first tool is that is very important after the checksheet is histogram. To get the histogram you need frequency distribution.</p><p><strong>What is frequency distribution?</strong></p><p>Frequency distribution is a table in which contains data arranged into classes with the number of objects that fall into each class. The number of observations that fall into each class is called frequency. But the frequency table itself will not give you any pattern, so you need a histogram.</p><p>You can think of histogram as a photograph of frequency distribution, It gives a pattern — whether the distribution is a bell curve skewed, right skewed, whether distribution has peaks, so on and so forth.</p><p>In statistics people always talk about the bell curve.</p><p>The bell curve is extremely useful for establishing standards in performance measures. therefore</p><p>bell curve is becoming a prerequisite for many applications.</p><p><strong>Uses of Histogram :</strong></p><p>1. In materials management for assessing materials strength.</p><p>2. We use histogram for estimating process capability. In context of manufacturing, it means is my process capable of meeting customer specification.</p><p>3. If it is not bell curve, it tell you corrective actions must be taken, because it indicates necessity of corrective actions, then after the corrective action is the system or process stable? Histogram will tell you.</p><p>4. Estimating machine capability in terms of speed and performance</p><p>5. We can compare different vendors, products, regions and performance across different locations using histograms.</p><p>To summarize histogram is a pattern recognition tool that tells you the shape of distribution and associated properties which help decision makers take remedial actions.</p><p>Let’s now move to statistical summary measures called statistical averages</p><p><strong>Statistical Averages</strong></p><p>When I say average, lot of people usually imply arithmetic mean, but that’s not correct.</p><p>There are many statistical averages — Mean, median, mode, harmonic mean, geometric mean etc.</p><p>We use a term here “Measure of Central Tendency”.</p><p>What is central about it? What is tendency ?</p><p>Whenever you take measurement of same characteristics a number of times and you form a dataset.</p><p>Then we will ask a fundamental question — Is it possible for me to pick up on average out of these measurements in such a manner that remaining items in the data have tendency to be around this value. Such a measure is called measure of central tendency.</p><p>Most often used measures of central tendency are</p><p>1. Arithmetic Mean</p><p>2. Median</p><p>3. Mode</p><p>Important thing to note is that all the three measures must be understood in the context of application. Context is extremely important.</p><p>I am assuming readers of this blog have basic idea of what they are and how they are computed.</p><p><strong>1. Arithmetic Mean:</strong></p><p>Arithmetic Mean is affected by extreme values or fluctuations in sampling. It is not the best average to use when dataset contains extreme values. (Very high or very low values)</p><p>For example let us consider CTC of 11 students ( in lakhs per annum)</p><p>3, 4, 6, 4, 7,5, 4, 5, 6, 18 , 5</p><p>Average = 6</p><p>But if Iremove the student who grabbed 18 lpa from the dataset, average is reduced to 4</p><p><strong>2. Median:</strong></p><p>It is the middle value when dataset is arranged in ascending ( or descending ) order of magnitude.</p><p>So where do we use median?</p><p>In market research,consumer preferences study, we always have a consumer ranking ( ordinal ranking). We use median as a reference point in the distribution pattern.</p><p><strong>3. Mode</strong></p><p>Mode is defined as that value that occurs most often. Mode also has resistant to outliers next to median. In Fact both mode and median are capable of fighting outliers better than mean.</p><p>When do we use mode?</p><p>For example, say you go to buy a pair of shoes at Nike, Bata or any shoe company’s outlet. You will notice that most frequently occurring size is always available in the inventory.</p><p>Caution : In few problems in real life, there will be more than one mode such as bimodal and multi modal values. In these cases mode cannot be uniquely determined.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2d570d41e50f" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>