Handling Class Imbalance in Online Transaction Fraud Detection

With the rise of internet facilities, a greater number people have started doing online transactions at an exponential rate in recent years as transaction system has eliminated need going to bank physically for every transaction. However, fraud cases also increased causing loss money consumers. Hence, effective detection is hour which can detect fraudulent automatically real-time. Generally, genuine are large than leads class imbalance problem. In this research work, using deep learning been proposed handle problem by applying algorithm-level methods modify model focus more on minority i.e., transactions. A novel function named Weighted Hard- Reduced Focal Loss (WH-RFL) achieved maximum True Positive Rate (TPR) cost misclassification few high TPR preferred over Negative (TNR) and same demonstrated three publicly available imbalanced transactional datasets. Also, Thresholding applied optimize decision threshold cross-validation frauds it experimental results that selection right thresholding method with yields better results.