Analysis of Fraud Detection Solutions Using Machine Learning (DSR Approach)

Abstract

The research aims to identify the problems and details of fraud detection methods in bank transactions using machine algorithms and to provide solutions in this field. Qualitative research is conducted for this aim, and the main problems are identified by reviewing previous research. Then, solutions are presented using the Design Science Research Methodology (DSR). The main topics identified from previous research include Data limitation, labeled data, Discovering new fraud patterns, Bias and Costs, and responsibility for false prediction. The proposed research model has been designed using results from previous studies and experts' opinions in this field. Using both supervised and unsupervised algorithms in the transaction registration process, labeling data based on discovered patterns, obtaining customer confirmation in cases where the system detects fraud, training, and continuous improvement of learning models using the generated data online are among the solutions of the suggested model. Also, it is suggested that the issue of reducing the error of false harmful data in the fraud detection process be investigated in future research.