Evaluating Trade-offs Between Error Rates in Machine Learning Credit Scoring Models

Several studies have explored the application of machine learning in credit scoring, however there is limited research focusing on the implication of the trade-offs between the false positive and false negative rates in these models. Trade-off occurs when one error is prioritized over the other and have an impact on both the lenders and borrowers. An increased false positive rate will misclassify more potential customers as high-risk, causing financial losses for lenders and negatively impacting credit applicants, while an increased false negative rate leads to missed opportunities for approving creditworthy applicants. This study assesses the trade-offs in the following machine learning models for credit scoring: logistic regression, multilayer perceptron, support vector machine and random forest trained on the German Credit and the Kenyan Uwezo Fund datasets and further tests a number of methods to control it so as to arrive at the trade-off that works best for the intended users. The outcome of the study showed that assigning equal cost to both errors, balancing class distributions through resampling techniques and adjusting thresholds affected the trade-offs between false positive rates and false negative rates in the models therefore managing the trade-offs towards an optimal point. The results of the study highlights the potential impacts on fairness and decision-making in credit scoring.