A Review of Machine Learning Models for Fraud Detection in Financial Systems

Abstract

In today’s banking on digital and payment ecosystem in electronic because of the huge fire growth in online dealings, payment on mobile and agreement systems in real-time. Knowledge of machine using ML provides a volition by learning fraud trends and patterns from history through behavioral attributes. This paper elaborate an analysis of machine knowledge models in a structured manner for discovery of fraud in systems which involves supervising methods, for example Support Vector Machines, Logistic Regression and Random Forest. On the other side, clustering, Isolation forest and One-class SVM approach towards Unsupervised and semi-supervised methodologies. Here Autoencoders with cases where labels are reviewed for fraud which are scare. Artificial Neural Networks (ANN) and Long Short-Term Memory networks are reviewed and examined for detection of fraud in sequential manner and for modelling temporal behavior. This paper discuss with the operational challenges and key research challenges which are concept drift, interpretability, severe class imbalance, privacy, constraints detecting in real-time and adaption which adversarial by fraudsters. Using metrics on performance such as model-wise evaluation and discussions, Comparative analysis are performed. Concluding with the idea of this paper which emerge directions which involves federated learning, online learning method, graph neural networks and explainable AI enhance detection of robustness in frauds in the real time scenario and financial environments.