Recent Progress on Financial Risk Detection in the Context of Transaction Fraud Based on Machine Learning Algorithms

Transaction Fraud, a type of financial operational risk, remains a major threat to financial sectors and continuously imposes devastating financial impacts. This study comprehensively reviews 41 cutting-edge publications on financial transaction fraud detection using Machine Learning from January 2018 to October 2025. We establish a taxonomy to categorize the selected work into four themes: Traditional Machine Learning, Deep Learning, Ensemble Method, and Hybrid Method. Each theme is evaluated in-depth, from strengths to weaknesses. Ensemble exhibits better performance over other methods with a recall of 92.7%, a precision of 96% and an F1-score of 92.66% on average, while Traditional ML ranks last in terms of average F1-score. Preprocessing strategies, like data balancing, can enhance performance, while feature engineering requires careful evaluation before implementation. Significantly, we assess financial implications, suggesting it is essential to integrate financial metric design, feature explanation, time series patterns, and data privacy considerations into financial fraud detection—a focus that aligns with risk management frameworks and regulations. By revealing current research gaps and suggesting future directions, our study provides practical guidance for researchers and practitioners to advance financial fraud detection strategies within a highly intricate financial ecosystem.