Online Credit Card Fraud Detection Using Bi-Strategy based Artificial Hummingbird Algorithm with Least Square Support Vector Machine

With the development of online payment and e-commerce platforms, the number of fraudulent transactions has increased significantly.The existing techniques do not perform hyperparameter tuning, which leads to suboptimal performance because the model is not optimized to effectively differentiate between fraudulent and legitimate transactions.This research proposes a Bi-Strategy-based Artificial Hummingbird Algorithm (Bi-AHA) with a Least Square Support Vector Machine (LSSVM) for fraud detection.The regularization in LSSVM helps by penalizing complex models, thereby preventing overfitting and enhancing generalization by controlling model complexity.The Bi-Strategy includes the Chebyshev Chaotic Map (CCM) and Random Opposition-based Learning (ROL), which are introduced into the traditional AHA to optimize hyperparameters of LSSVM for obtaining optimal global search, enhancing convergence, and avoiding local optima.By using hyperparameter tuning, the model improves its capability to detect credit card frauds in online transactions.The Credit Card Fraud (CCF) dataset is preprocessed using min-max normalization and Adaptive Synthetic (ADASYN) sampling technique is applied for data imbalance.The Bi-AHA-LSSVM obtains 99.98% accuracy, 99.91% precision, 99.86% recall, 99.97% specificity, 99.88% f1-score, and 99.56% Area Under Curve (AUC) for the CCF dataset when compared to Auto Encoder SVM (AE-SVM).