Using Memory-Based Reasoning For Predicting Default Rates On Consumer Loans

Abstract

In recent years, financial institutions have struggled with high default rates for consumer lending. An ability to reliably predict the probability of consumer loan defaults would have a significant impact of the profitability of that lending for these institutions. In response to this need, the financial institutions have employed loan analysis techniques such as logistic regression, discriminant analysis, and various machine learning techniques to improve the accuracy of detecting loan defaults.  The objective of these techniques is to more precisely identify creditworthy applicants who are granted credit, thereby increasing profits, from non-creditworthy applicants who would be then denied credit, thus decreasing losses. The objective of this article is to employ an emergent data analysis technique, memory-based or case-based reasoning method, to this problem to test its accuracy in discriminating between good and bad loans. This paper examines historical data from consumer loans issued by a financial institution to individuals that the financial institution considered to be qualified customers.  The data set consists of the financial attributes of each customer and includes a mixture of loans that the customers paid off or defaulted upon. The paper then compares the performance of this technique to other data mining techniques proposed in earlier works and analyzes the risk of default inherent in each loan for each technique.