Comparative Evaluation of Predictive Modeling Techniques on Credit Card Data

Abstract

Credit Scoring studies are very important for any financial house. Both traditional statistical and modern data mining/machine learning tools have been evaluated in the credit scoring problem. But very few of the studies facilitate the comparison of majority of the commonly employed tools in single comprehensive study. All the tools such as LDA (Linear Discriminant Analysis), SVM (support vector machines), Kernel density estimation, LR (logistic regression), GP(genetic programming), K neighborhood, which are available in SAS enterprise miner 6.2. The results revealed that support vector machine and genetic programming are superior tools for the purpose of classifying the loan applicant as their misclassification rates were least as compared to others. whether to grant credit or not. The precise judgment of the creditworthiness of applicants allows financial institutions to increase the volume of granted credit while minimizing possible losses. The increasing number of potential applicants has driven the need for the development of sophisticated fraud detection techniques that automate the credit approval procedure. Earlier credit scoring was restricted to statistical techniques such as discrimination between several groups in a data sample. History reveals that credit scoring came into existence in early thirties of the nineteenth century when some financial institution decided to classify their applicants with respect to default status. In late 1950's the first automated system was used to develop predictive model of the applicants based on their historical data. After the card holders default, several data collection procedures are undertaken. These procedures are expensive relative to the size of most loans. They are often useless when the card holder has gone bankrupt. Therefore, it is significant for the card issuers to identify card holder type at early stage in order to minimize lending to risky customers before the default occurs. It means that it becomes necessary to maximize the 'True positives' (TP). The true positive rate (TRP) is known as sensitivity and the true negative rate (TNR) is sometimes called specificity.