Cost-sensitive Risk Induced Bayesian Inference Bagging (RIBIB) for credit card fraud detection

Credit card fraud represents one of the biggest threats for organizations due to the probability of huge losses associated with them. This paper presents a cost-sensitive Risk Induced Bayesian Inference Bagging model, RIBIB, for credit card fraud detection. RIBIB proposes a novel bagging architecture incorporating a constrained bag creation method, a Risk Induced Bayesian Inference method as a base learner and a cost-sensitive weighted voting combiner. Experiments on Brazilian Bank data indicate 1.04-1.5 times reduced cost. Experiments on UCSD-FICO data exhibit robustness of the model in handling unseen data without any need for domain specific parameter fine-tuning.