DEVELOPMENT OF A GENETIC ADAPTIVE NEURO-FUZZY MODEL FOR LOAN RISK PREDICTION

Abstract

Loan assessment is an important issue that banks and other financial institutions must deal with because of the risk involved. It is essential to evaluate if a loan applicant will default or not in order to minimize loss. In time past, loan applicants are evaluated based on expert judgment and the use of statistical techniques, these approaches have been found to be inefficient at loan risk evaluation. With the ever increasing rate of amount of loans to be considered for approval every day and the rate at which borrowers default, accurate prediction of loan risk is indispensable to lending organizations. In this research, factors that are relevant for loan risk are first determined using the wrapper and filter methods. Then a Genetic Adaptive Neuro-Fuzzy System (GA-ANFIS) is developed for Loan Risk Prediction. The system is tested on German credit dataset as a case study. The data set has 1000 instances each of which has 24 attributes. The dataset is divided into a training set (700 instances), which was used to train the model and testing set (300 instances), which was used to test the model. The GA-ANFIS on the features selected by wrapper method achieved a recognition accuracy of 79.33% while GA-ANFIS on the features selected by the filter method achieved a recognition accuracy of 77.0 %. A comparative analysis of the model with some other models such as logistic regression, decision trees and neural networks shows that the proposed model outperforms the other models on some standard metrics and underperforms on some of the standard metrics.