Applying Machine Learning to enhance payments systems security

Abstract

During the last two decades, the economic losses because fraudulent card payment transactions have tripled. The significant percentage of losses is because of fraud on e-commerce transactions. Nowadays, there is a clear trend to use more and more mobile devices to make electronic purchases, and it is estimated that this trend will continue in the coming years. In the card payment scheme, big financial institutions process millions of transactions every day; thus, they can model the processed transactions to predict fraud. On the other hand, merchants process a much lower number of transactions, but they have access to valuable information that they can collect from the devices that users utilise during the transaction. In this thesis, we propose a series of measures to enhance the security of these two scenarios based on past transactional data and information collected from the users’ device. Most of the approaches proposed so far to model processed transactions were based on supervised Machine Learning techniques. We propose a fraud detection system for card payments based on an unsupervised machine learning technique; thus, the system may be able to recognise new patterns of fraud. On the other hand, we are looking far ahead, and because of the increment of use of mobile devices to conduct payments, we propose a series of measures to enhance the security of the mobile payment system. We have proposed a user identification and verification systems for smartphones. We base the identification and verification systems on motion data, so the systems will not require any explicit action from users.