Comparison of Accuracy of Support Vector Machine Model and Logistic Regression Model in Predicting Individual Loan Defaults

摘要

Prediction of loan defaults is critical to financial institutions in order to minimize losses from loan non-payments. Some of the models that have been used to predict loan default include logistic regression models, linear discriminant analysis models and extreme value theory models. These models are parametric in nature thus they assume that the response being investigated takes a particular functional form. However, there is a possibility that the functional form used to estimate the response is very different from the actual functional form of the response. In such a case, the resulting model will be inaccurate. Support vector machine is non-parametric and does not take any prior assumption of the functional form of the data. The purpose of this study was to compare prediction of individual loan defaults in Kenya using support vector machine and logistic regression models. The data was obtained from equity bank for the period between 2006 and 2016. A sample of 1000 loan applicants whose loans had been approved was used. The variables considered were credit history, purpose of the loan, loan amount, saving account status, employment status, gender, age, security and area of residence. The data was split into training and test data. The train data was used to train the logistic regression and support vector machine models. The study fitted logistic regression and support vector machine models. Logistic regression model showed an accuracy of 0.7727 with the train data and 0.7333 with test data. The logistic regression model showed precision of 0.8440 and 0.8244 with the train and test data. The SVM (linear kernel) model showed an accuracy of 0.8829 and 0.8612 with the train and test respectively. The SVM (linear kernel) showed a precision of 0.8785 with the train data and 0.7831 with the test data. The results showed that support vector machine model performed better than logistic regression model. The study recommended the use of support vector machines in loan default prediction in financial institutions.