Risk Management Model in Electronic Banking Using Machine Learning Based Fraud Detection System

Risk management techniques have been used for a long time, they are just now becoming well-known in the electronic banking industry. At this stage, risk management is essential to the entire operations of financial institutions. Financial institutions must evaluate the procedures that support their business functions against prospective risks, weaknesses, and possible effects to determine the degree of risk exposures and any lingering dangers. Additionally, high financial processing expenses are necessary due to the notable rise in fraud in e-banking. Consequently, fraud detection in e-banking has emerged as an intriguing subject for risk management. The application of category weight-tuning parameters to regulate the value of authentic and fraud transactions is examined in this paper. By considering the voting procedure, this study suggests using weight-tuning as a pre-process for imbalanced datasets in addition to Cat-Boost and XG-Boost to enhance the Light-GBM technique's efficiency. Lastly, this study employs deep learning to adjust the hyperparameters, especially the suggested weight-tuning to significantly enhance efficiency. This study employs recall-precision measurements to the conventional ROC-AUC to more effectively handle unbalanced databases. A five-fold cross-validating technique is used for assessing Cat-Boost, Light-GBM, and XG-Boost independently. Additionally, the efficiency of the paired methods is evaluated using the predominant vote ensemble learning technique. The outcomes show that Light-GBM and XG-Boost meet the optimal degree criterion of ROC-AUC = 0.952, precision 0.79, recall 0.80, Fl-value 0.79, and MCC 0.79. When compared to the state-of-the-art techniques, this is a major advance.