A Hybrid Machine Learning Framework for E-commerce Fraud Detection
We currently see a large increase in e-commerce sector; it is becoming a central trend in the banking industry. Fraudsters keep up with modern technologies, and use weak points in human psychology and security systems to steal money from regular users. To ensure the required level of security, banks began to apply artificial intelligence in their anti-fraud systems. Fraud detection can be formulated as a classification problem with a case-based reasoning or knowledge extraction task with unbalanced classes. In this paper we present a framework of models based on various approaches of artificial intelligence, such as neural networks, decision trees, copula models and others to recognize payment behavior of fraudster. The considered framework is evaluated with different metrics and implemented in an actual anti-fraud system, which leads to an improvement of the system performance. Finally, the interrelation between the anti-fraud system indicators and banks operational risks is discussed in this paper.