Fraud Detection Models and their Explanations for a Buy-Now-Pay-Later Application

Buy-now-pay-later (BNPL) has been increasingly adopted by young shoppers as well as older generations because applying for a BNPL service is hassle-free compared to credit card applications. However, preventing fraudulent transactions is indispensable. In this study, machine learning techniques have been used to create fraud detection models for a BNPL company. Of the four algorithms tested, Random Forest was the top performer, with an impressive F1 score of 0.92 and a low false positive rate of 0.072. To understand how decisions are made by the random forest classifier and build human trust, we use SHAP (SHapley Additive exPlanations), a popular explainable artificial intelligence (XAI) method, to interpret the model with both global and local explanations. It is discovered that the detection rules used by the company’s human experts involve features deemed important by SHAP. Furthermore, we discovered other features that are useful for determining whether a transaction is fraudulent or not, which human experts have not considered before. Through XAI explanations, humans can work together with prediction models to detect fraudulent transactions and learn from each other.