<?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 Yasir Salim on Medium]]></title>
        <description><![CDATA[Stories by Yasir Salim on Medium]]></description>
        <link>https://medium.com/@yasirsalim?source=rss-336fd42f6fc1------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*zUSgO4fDsrssBka7.</url>
            <title>Stories by Yasir Salim on Medium</title>
            <link>https://medium.com/@yasirsalim?source=rss-336fd42f6fc1------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 15:13:35 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@yasirsalim/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[Classification Algorithms and Social Credit Scoring]]></title>
            <link>https://medium.com/@yasirsalim/classification-algorithms-and-social-credit-scoring-aa4bd1afbdd?source=rss-336fd42f6fc1------2</link>
            <guid isPermaLink="false">https://medium.com/p/aa4bd1afbdd</guid>
            <category><![CDATA[logistic-regression]]></category>
            <category><![CDATA[classifier-algorithms]]></category>
            <category><![CDATA[credit-scoring]]></category>
            <dc:creator><![CDATA[Yasir Salim]]></dc:creator>
            <pubDate>Sat, 28 Jul 2018 14:19:30 GMT</pubDate>
            <atom:updated>2018-07-30T11:18:59.613Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*B9W-VKogWch4L5LDfjGieQ.jpeg" /><figcaption>Social Credit Scoring</figcaption></figure><h3>Abstract:</h3><p>This article presents an approach to predict the social credit scores of customers using the Logistic Regression machine learning algorithm. The research objective of this project is to understand social credit scoring and classifier algorithms and to perform a comparative study between feature selection and feature extraction using the Logistic Regression machine learning algorithm. In this article I used eleven features to predict credit score and then I did analysis of logistic regression algorithm in credit scoring.</p><h3>Key Words</h3><p>Social credit scoring, Classifier Algorithms, Logistic Regression Algorithm</p><h3><em>Introduction</em></h3><p>As soon as human beings started to cooperate and create a community, there are no doubts that they started to borrow and repay as well. Main concern of lenders is that the loan should be granted only to those customers, who are going to return the loan back. But how is the lender able to reveal the nature of a potential customer. Until 1970, traditional credit assessment was relied on “gut feel”. Which means that If manager is satisfy then bank will give loan otherwise they will not give loan and the process was also very slow. Then bank started credit risk assessment to enhance the procedure of loan. With rapid growth of the credit industry in last few decades, credit evaluation of loan applicants become an important issue. The credit scoring or credit evaluation model is simply classify loan applicants into two categories: “Good Loaner” and “Bad Loaner” with the help of classifier algorithms. Classifier algorithms like decision tree, logistic regression and K-nearest neighbor are used in credit scoring. In these days Credit scoring has a new face in the form of social credit scoring. In social credit score module, companies are not only analyzing financial status of person but also analyzing the social life of person as well. Chinese government’s and other companies like Alibaba and ANT attempted to create a social credit system aimed at rating the trustworthiness of people and companies.</p><h3>Discussion</h3><h3>Social Credit Scoring</h3><p>In this era many of your daily activities are constantly evaluated and monitored: what you buy at the shopping or online, what is your location; your social circle and your interaction with them and what bills and taxes you pay. Monitoring of these things is not difficult because data-collecting behemoths like Google, Facebook and Instagram or health-tracking apps such as Fitbit are doing this. Different data analyst companies like ANT are using this data to assign the score to every individual.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/610/1*VUuUUfqt5XueBo7OfPu4KA.png" /></figure><p>And the purpose of this score is the same as in bank credit assessment to analyse trustworthiness of individual. This is happening in China under the supervision of government and Alibaba. Here are the attributes of social credit scoring.</p><p><strong>Credit History</strong></p><p>Credit history is the past payment history and indebtedness. This consists of areas such as credit card repayment, utility bill payments and ratings on e-commerce website such as Taobao.</p><p><strong>Behavior &amp; Preference</strong></p><p>Behavior user’s including how active he/she is, what websites he/she visits, the product categories he/she shops at and the consumer segment the user belongs to, etc.</p><p><strong>Fulfillment Capacity</strong></p><p>reflects the user’s ability of adhering to contracts such as financial products and loans. It is evaluated based on asset information, including payment for social insurances , properties and cars, etc. Indicators also include use of Alibaba’s financial products / services and Alipay account balances.</p><p><strong>Identity Characteristics</strong></p><p>rates the extent and accuracy of user’s personal information. For example, home address, length of time of residence, mobile phone number, education level and employment. User can fill in information or import information from LinkedIn or other external systems.</p><p><strong>Social Relationship</strong></p><p>reveals the user’s influence within his/her network, his/her influence to the network and his/her friends’ credit scores.</p><p>Every attribute has certain weight-age in scoring and in below picture their are the weights of attributes.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/203/1*uxBgc2lvXSSdKkBoqEW1Mw.png" /><figcaption>Credit Scoring division with their weights of attributes</figcaption></figure><p>With the help of these weight the distribution of credit score become easy. So here is the second picture which distribute the credit score to every attribute.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/490/1*WezsxMz9qg-_4PP-j48ShQ.png" /><figcaption>Credit Score according to the attributes weithages</figcaption></figure><p>By using different classifier algorithms we can predict the score and then by using this score we can predict that either the individual is defaulter or not.</p><h3>Classifier Algorithms</h3><p>There are many classifier algorithms which help in credit scoring and I will be discussing three of them which are given below.</p><ol><li>Logistic Regression</li><li>Decisions Tree</li><li>K-Nearest Neighbor</li></ol><p><strong>Logistic Regression:</strong></p><p>Logistic regression is a machine learning algorithm for classification. In this algorithm, the probabilities describing the possible outcomes of a single trial are modeled using a logistic function. Methods for creating credit scorecards can be divided into two groups: statistical and non-statistical. Until the 1980, statistical methods like linear regression models were used to create decisions. Statistical analysis is the general approach for credit assessment.</p><p>In logistic regression the response variable y ∈ {0,1} follows a Bernoulli distribution. The response variable reﬂects the creditworthy of the customer:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/392/1*Xkb_jmq_dDC1LjNlWBq2xw.png" /></figure><p><strong>Decision Tree</strong></p><p>Given a data of attributes together with its classes, a decision tree produces a sequence of rules that can be used to classify the data. A decision tree consists of a set of sequential binary splits of the data. The algorithm begins with a root node containing a sample of good and bad credit applicants.Then, the algorithm loops over all possible binary splits in order to ﬁnd the attribute x and corresponding cutoﬀ value c which gives the best separation into one side having mostly good credits and the other mostly bad credits.</p><p><strong>K-Nearest Neighbors</strong></p><p>Neighbors based classification is a type of lazy learning as it does not attempt to construct a general internal model, but simply stores instances of the training data. Classification is computed from a simple majority vote of the k nearest neighbors of each point.</p><h3>Credit Scoring Using Logistic Regression</h3><p>The main aim is to use a classifier to do social credit risk prediction. In general classification tasks require three major steps:</p><p>1. Data preprocessing (such as removing outliers)</p><p>2. Feature Engineering</p><p>3. Training classification algorithm and predictions (classifier in this case is logistic regression)</p><p><strong>Data Pre-Processing</strong></p><p>In this example we have a total 10 specific features for each row in the dataset and the 11th feature is actually what we want to predict, in this case that is the serious delinquency in 2 years. It’s a binary variable which is 0 if a person has not paid his loans properly in the last 2 years and 1 if he is a good customer and repays on time.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/914/1*32pwfHWIQ6TkFzXkCOdStA.png" /></figure><p>The other variable may or may not be useful but in this step we don’t care about their relevance but rather we want preprocess or clean the data. In this example, preprocessing involves removing rows with empty values, removing any data that seems like outliers, removing invalid age values such as age = 0, etc. This step is necessary as it gets rid of unnecessary data and helps us in the feature engineering step.</p><p><strong>Feature Engineering</strong></p><p>Feature engineering can be loosely divided into two categories feature selection and feature extraction. In case of feature selection, we analyze the given data and select the most relevant features for classification. In feature extraction we further process the data to reduce its number of features or dimensionality by using other algorithms such as PCA. The main difference between these two approaches is that while feature selection allows us to understand what variables are being used to classify, whereas feature extraction may transform the features into incomprehensible numeric data. Either way the classifier doesn’t care as long as it is getting a constant stream of numbers to train on. In this case we will use feature selection as we want to know what variables are important to understand to determine a person’s credibility.</p><p>In order to select features after removing outliers we can use a number of techniques to find the correlation between features and our feature to be predicted. In this case, a heat map of correlation between all the features is used. It can be seen 4 features seem to be correlated to our target feature. So we create a subset of these 4 features. We can also try selecting more features and see if our prediction accuracy increases or not. After selecting features its time to finally train the classifier.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/810/1*geghJzrNdRdivJSqllSIPA.png" /><figcaption>Heat map of correlation between all the features</figcaption></figure><p><strong>Training classification algorithm and prediction</strong></p><p>In this case we will use logistic regression on the 4 data feature set. Using this algorithm with the selected features we get the following result.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/702/1*pLRymO2Wx94JWleqRZ8Wag.png" /></figure><p>Increasing the number of features does not help though as evident by the result. You can also try improving this classification by either using features which are correlated or using another classification algorithm such as linear SVM.</p><h3>Conclusion</h3><p>The purpose of this work was to understand the social credit scoring and the contribution of classifier algorithms in credit scoring. As we saw that credit scoring is becoming the new trend to analyse the financial standing of individual and on the basis of these scoring, decisions would be take place about the loan, travelling and the class of every individual and the second thing was that how classifier algorithms helps to predict the score trustworthiness and standing of individual. For this we did Data preprocessing and then Feature engineering where we did feature selection and feature extraction and at the end I applied classifier algorithm to find out the credit score.</p><h3>References</h3><ol><li>CREDIT RISK SCORECARD ESTIMATION BY LOGISTIC<br>REGRESSION by Aleksandr Peussa</li><li>Social Media-Driven Credit Scoring: the Predictive Value of Social Structures by Tianhui Tan and Tuan Q. Phan</li><li><a href="https://www.wired.com/story/age-of-social-credit/">https://www.wired.com/story/age-of-social-credit/</a></li><li><a href="https://www.wired.co.uk/article/chinese-government-social-credit-score-privacy-invasion">https://www.wired.co.uk/article/chinese-government-social-credit-score-privacy-invasion</a></li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=aa4bd1afbdd" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>