Tree-Based Methods for Loan Approval

Loan approval is one of the most important processes that any banking organization owns. The acceptance or rejection of any loan application has a direct impact on the bank revenue and the profitability in quarterly issued financial statements. Though loan approval is a critical process, the actual decision made is not a straightforward procedure and comes with a lot of uncertainties. Recently, statisticians and data scientists have tried to automate this process to minimize risk and increase profitability by applying different statistical learning methods. In this work we explore a framework with an application by applying tree-based methods on publicly available dataset. This work aimed at developing a high performance predictive model for loan approval prediction using decision trees. Experiments were made in different varieties of tree methods ranging from the most simplified and comprehensible decision tree reaching up to the most complex random forests. Results yielded inadequate performance with respect to simplified decision trees due to the highlight correlated and complex feature space, majority of critical parameters affecting loan approval was not reflected upon and yielded an impractically over-simplified tree. However, boosting came in superior in terms of performance, relevance and interpretation via the importance chart scoring accuracy on testing dataset [98.75%] specificity [100%], Minority class prediction accuracy [92.85%], and classification efficiency of [97.0%]. Therefore, boosting-based decision-tree predictive model was recommended to facilitate decision making regarding the eligibility of loan applicants based on their characteristics.