Machine Learning Approaches for Identifying Fraudulent Banking Transactions: A Financial Management Perspective

Online banking, credit card purchases, and mobile wallet are all examples of the growing popularity of banking activities. Financial organisations, retailers, and consumers all face the formidable problem of detecting fraudulent financial transactions. The traditional rule-based detection methods are frequently insufficient due to the growing sophistication of fraudulent activities. The development of AI has made it possible to intelligently analyse massive amounts of financial data using machine-learning-based algorithms in order to spot fraudulent activities. Recent studies have shown that ML approaches may effectively identify fraudulent transactions in massive volumes of payment data, which is a crucial component of cyber-crime organisations' operations. This research provides machine learning models to detect fraudulent financial transactions using the PaySim dataset. Label encoding, min-max scaling, and SMOTE-based balancing are utilized to overcome the intrinsic class imbalance in fraud detection tasks. Classification models like SVM, NB, RF, and LR are assessed using F1-score, recall, accuracy, and precision. The outcomes indicate that Logistic Regression excels in accuracy with 98.99%, whereas Random Forest excels in recall and Naïve Bayes excels in precision. The study emphasizes the need for strong fraud detection systems and the trade-offs involved in optimizing evaluation measures to combat evolving financial crimes.