ANOMALY DETECTION AND RISKS MANAGEMENT IN ONLINE PAYMENT SYSTEM

ABSTRACT

The introduction and adoption of Internet in businesses has been a major breakthrough in online transactions, eliminating the problems of physical transactions using cash. Due to the anonymous nature of the internet and its increasing use for online transactions has led to increase in number of fraudulent transactions recorded. This comes with great risk, as the number reported fraud are almost on daily bases and includes credit card fraud, phishing, and business compromise with huge financial losses running into billions of dollars.Efforts in curbing fraud related cases in e-commerce are still ongoing and very massive. The need to mitigate growing and prevalent cases of e-commerce related frauds is of great importance, hence led to the conception and development of this paper. This project therefore aims at developing an anomaly detection system for e-commerce, using machine learning technique. The system detects abnormal transaction at real time in an online platform, with the aim of mitigating the risk involved in online transactions. For this project, a model is built using Isolation Forest algorithm, and is trained using dataset containing transactions of card holders, within a specified period of time. A total of 248,800 records of user transactions with 23 corresponding attributes or features of the dataset which includes time elapsed between transactions, amount, and target status of transaction. The anomaly detection model was design using python programming language, and was implemented in a Jupiter Development Environment.