Comparison of discriminant analysis and logistic regression for the classification of borrowers of microfinance institutions

Abstract

This paper uses logistic regression and discriminant analysis to predict default in the loan portfolio of Microfinance institutions. Out of the 723 borrowers the logistic regression correctly classified 92.7% of good borrowers and 73.1% of bad borrowers. As for the discriminant analysis model it correctly classified 94.2% of the good borrowers and 84.5% of bad borrowers. The overall predictive accuracy of both methods was good enough with the logistic regression having the highest value of 81.6%. The results suggest that both methods can be used in selecting borrowers in the loan portfolio of microfinance institutions. Nevertheless given that the discriminant analysis does not meet up with its underlying assumptions, logistic regression is recommended.