Using Supervised Machine Learning Approaches To Detect Fraud In The Banking Transaction Network

Nowadays, criminal frauds occur in an organized manner in the banking sector. This issue is challenging since the number of organized frauds associated with such areas is estimated to range from 2% to 5% of the global gross domestic product (GDP). The people committing organized fraud use Internet-based financial services and conventional financial services. Accordingly, they use more complex plans and maps to avoid being recognized through organized fraud fighting systems. Due to the complexity and variety of fraud methods, the transaction may not seem suspicious initially. Hence, it is crucial to consider the interactions between the cards. For this purpose, the use of network theory is recommended. The current paper aims to classify each transaction as illegal or legal correctly. Therefore, extensive data analysis is used to organized fraud in the bank transaction network. Besides, a comparison between supervised learning algorithms is presented on a dataset with 46,316 transactions related to customers' card activities to distinguish between illegal and legal transactions. According to the Accuracy, Precision, Recall, and F1-Score criteria values, random forest and XGBoost could be considered suitable predictive models for fraud detection.