Automated Financial Fraud Detection in Banking Systems based on Stacking Ensemble Model

In recent scenarios, the financial fraud detection in banking systems involved in identifying and preventing fraudulent activities like credit card fraud and account takeovers. However, effective detection is required for analyzing customer behavior and transaction patterns to prevent financial losses. Moreover, the existing Boosting techniques struggled with the underlying patterns which change over time. Therefore, this research proposes Stacking Ensemble Model (SEM) for effective financial fraud detection and handles the drift patterns with the diversity of base models. Initially, fraudulent data is collected from Credit Card Fraud Detection (CCFD) dataset which is openly accessible at Kaggle. Then, input data undergo preprocessing with sampling techniques to balance data and Z-score normalization to standardize data which is done by subtracting mean and dividing by the standard deviation for each feature. After that, the preprocessed features are selected with the help of Correlation Analysis (CA) to select the optimal features. Finally, the proposed SEM model is incorporated to detect financial fraud by fusing the predictions from diverse base models. From the results, the proposed SEM model outperformed existing Boosting Techniques in terms of accuracy (99.8%), Area Under Curve (AUC) as (93.5%) respectively.