Attention Based Isolation Forest Integrated Ensemble Machine Learning Algorithm for Financial Fraud Detection

Online card transactions and mobile payment services made easy access for the users to facilitate the payment throughout the world. This attracted the interest of the fraudsters and observed increased cases of fraudulent activities. Many Machine Learning (ML) methods were used to design the Financial Fraud Detection (FFD) system, however extreme class imbalance and high computational complexity negatively affected the performance of the system. The aim of current research is to integrate the Attention Based Isolated Forest (ABIF) with Ensemble Machine Learning (EML) algorithm which includes Random Forest (RF) and AdaBoost (AB). The Paysim financial credit card data is initially pre-processed to perform data scaling and remove the duplicates. The data is sampled using Synthetic Minority Over-sampling TEchnique (SMOTE) for solving data imbalance issues. The relevant features are extracted using Residual Network (ResNet) followed by the feature selection by ABIF method. The final EML based trained FFD model is used for detecting the frauds which resulted in the overall detection accuracy 99.76%, precision of 98.64%, recall of 98.02%, and f1-score of 97.45% which are superior when compared to the existing FFD models namely Random Under Sampling + eXtreme Gradient Boosting (RUS+XGB) and Logistic Regression- RF (LR-RF).