AI-Powered Fraud Detection in Payments Using Long-Term Behavior Sequence Modeling

AI-based fraud detection due to payment riskis gaining traction due to the advent of payment technologies.Many traditional fraud detection models cannot identify long-term behavior, making fraud detection inefficient. The proposedframework on fraud detection based on Deep Learning usesLong-term Payment Behaviour Sequence Folding (LBSF) whichanalysess sequential payment behaviour for detection of anomalieswhich show fraud behaviour. This fraud detection techniqueuses deep learning essentially composed of a recurrent neuralnetwork. We use Transformer-based models, such as Longformer,Informer, Reformer, Mamba, and BERT4Rec, to encode dataabout merchants, transaction frequency, and spending behaviouracross multiple fields. We build hierachical sequences at themerchants level by reordering payment behaviour. We improvefraud detection through relational learning. The frameworkis tested on a large-scale real-world payment dataset fromTencent Mobile Payment, demonstrating a better fraud detectionperformance than baseline methods. The results from theseexperiments show that including a model for long-term paymentbehavior helps to reduce false positives in the detection of high-risk transactions. The results show that AI is pretty successful atdetecting payment fraud. So, deep learning methods are superuseful for finance.