A Review of Financial Fraud Detection Methods Based on Machine Learning

Abstract

Financial fraud seriously reduces capital market efficiency and harms investors' interests. Traditional auditing methods also have clear limits. Based on stages of technological development, this paper divides the use of machine learning in financial fraud identification into three stages, from statistical learning and classical machine learning to deep learning. It reviews how machine learning technologies have been applied in financial fraud detection and examines the main principles, strengths, and limits of different methods. The study finds that ensemble learning and graph neural networks are effective in identifying related-party transactions and abnormal patterns, but they still face problems such as data imbalance and weak interpretability. This paper offers a reference point for studying intelligent financial fraud identification from the perspective of technological change.