Fraud Detection in Financial Transactions: A Machine Learning Approach vs. Rule-Based Systems

A comprehensive investigation of methods has been triggered as a result of the requirement to detect and prevent fraud in the dynamic environment of financial transactions, which is constantly developing. The purpose of this study is to investigate the effectiveness of Support Vector Machine –(SVM) and Rule-Based Systems (RBS) in identifying fraudulent activity. The ML technique achieves higher performance, with an F1-score of 90%, an accuracy score of 95%, a precision score of 92%, and a recall score of 89%. Having said that, this does come at the cost of higher computing needs, which results in a processing time of 120 milliseconds. RBS, on the other hand, exhibits faster computing efficiency (50 milliseconds), but at the expense of worse accuracy metrics. The Receiver Operating Characteristic (ROC) curve is a graphical representation that demonstrates the discriminatory capability of ML in locating fraudulent transactions. In addition, hybrid models are investigated, with an emphasis placed on a nuanced approach that strikes a compromise between accuracy, computational efficiency, and interpretability. Recommendations based on practical experience advocate utilising ML for accuracy and taking RBS into consideration for computing efficiency and interpretability. Hybrid models offer a good balance between these two approaches. This research makes an important contribution to the ongoing discussion concerning the enhancement of fraud detection systems in the financial sector by providing significant new insights.