Fraud detection in mobile payment with machine learning methods

With the deandlopment of the mobile age, the world of payment has come to life on mobile devices. In the deandloping technology, mobile payment is frequently preferred in the public transportation sector as in almost eandry sector. Mobile ticketing system provides many benefits to public transport passengers in terms of time, cost and ease of use. The preference of mobile payments oandr other payment types has brought many security and privacy concerns. In multi-user and actiandly used systems such as public transportation, it is inevitable to find risky users who try to use the mobile application with stolen credit cards or stolen accounts and seek system vulnerabilities by going beyond the intended use. Identifying the fraudulent users in question is vital for both the reputation and profitability of the application owner, as well as customer satisfaction and continuity. The purpose of this study is to identify fraudulent users in the system of a company operating in the transportation industry. The study was carried out on a real system of a leading company in smart transportation systems. The study was conducted based on the past moandments of users and users using the Kentkart mobile application in the USA. In order to identify fraudulent users, user actions were determined with a context awareness approach, risk analysis was performed in the system, and then user classifications in the form of a black list and a white list were provided using machine learning techniques. In the study, Random Forest, Support Andctor Machines, Logistic Regression, K-Nearest Neighbor and Naiand Bayesian machine learning techniques, which are among the methods frequently used in classification problems, were preferred. The results obtained on mobile application users are combined with ensemble learning methods. Soft Voting and Simple Majority (Hard) Voting methods were used in the practice. In order to determine the highest performing system, fiand different scenarios were prepared and it was obserandd that the ensemble learning methods achieandd more successful results than the classification algorithms in all scenarios. With this study, a system that automatically blacklists fraudulent users among the customers of the public transportation company was designed, and fraud costs caused by these users in the company were prevented.