Transfer Learning Strategies for Credit Card Fraud Detection

Credit card fraud jeopardizes the trust of customers in e-commerce transactions. This led recent years to major advances design automatic Fraud Detection Systems (FDS) able detect fraudulent transactions with short reaction time and high precision. Nevertheless, heterogeneous nature behavior makes it difficult tailor existing systems different contexts (e.g. new payment systems, countries and/or population segments). Given cost (research, prototype development, implementation production) designing data-driven FDSs, is crucial for transactional companies define procedures adapt pipelines challenges. From an AI/machine learning perspective, this known as problem transfer learning . paper discusses transfer approaches credit detection their assessment a real setting. The case study, based on six-month dataset (more than 200 million transactions) provided by industrial partner, relates models developed European country another country. In particular, we present discuss 15 techniques (ranging from naive baselines state-of-the-art approaches), making critical quantitative comparison terms precision scenarios. Our contributions are twofold: (i) show that accuracy many methods strongly dependent number labeled samples target domain (ii) propose ensemble solution self-supervised semi-supervised adaptation classifiers. thorough experimental shows both highly accurate hardly sensitive samples.