Leveraging AI and Machine Learning for Enhanced Fraud Detection in Digital Banking System: A Comparative Study

The merger of artificial intelligence (AI) and machine learning (ML) technologies has drastically changed the field of fraud detection in digital banking in recent years. The increasing frequency of digital transactions in contemporary financial systems has increased the risk of credit card fraud (CCF), necessitating the use of effective detection techniques. Despite the potential of machine learning (ML) technology, dataset imbalance limits the accuracy of fraud detection. This study examines how well data balancing strategies and ensemble learning approaches can identify CCF in unbalanced datasets. This study examines how well data balancing strategies and ensemble learning approaches can identify CCF in unbalanced datasets. In addition to evaluating data balancing strategies like SMOTE, ADASYN, and Random Oversampling, we examine four machine learning algorithms: KNN, Decision Tree, SVM, and LSTM. Significant gains were made by using balancing techniques: The greatest results are obtained with ROS on LSTM, with Recall 89.6%, F1-Score 91.5% and Precision 98.4%, With Precision 97.6%, Recall 91.5%, and F1-Score 93.9%, Random Oversampling ranks second on SVM, whereas SWOT on SVM is third with Precision 91.7%, Recall 86.3%, and F1-Score 89.9 %. The study's findings demonstrate that using SVM and LSTM classifiers in conjunction with oversampling strategies like Random Oversampling yields precise and unaffected results for identifying CCF in unbalanced datasets, assisting in the development of trustworthy fraud detection systems for practical uses.