ABSTRACT
176Large risks associated with the approval of bad loans has forced financial institutions such as banks to carefully choose their customers. In this chapter, a two-stage decision support system is developed to enable banks in establishing an efficient loan portfolio that incorporates their strategic goals. Classification of customers into two levels of credit risk (high risk or low risk) is addressed in the first stage. Five different machine learning techniques (logistic regression, random forests, neural networks, gradient boosting, and stacking) are developed to classify the customers, and the best technique is selected using different performance measures such as sensitivity, specificity, and accuracy. In the second stage, a multicriteria mathematical programming model is developed to obtain a diversified loan portfolio with the objective of achieving higher returns and lower risk. The functionality and applicability of the two-stage decision support system are shown using real-world data.
