Credit Card Fraud Detection Using Non-Overlapped Risk Based Bagging Ensemble (NRBE)

Fraud due to credit card misuse costs consumers several billions of dollars annually. This is due to the huge usage levels and inability of the systems to automatically detect the anomalies. This paper analyzes the implicit nature of data with noise and imbalance and proposes a Non-overlapped Risk based Bagged Ensemble model (NRBE) to handle imbalance and noise contained in the credit card transactions. The bagging model has been enhanced in terms of a novel bag creation model and an effective risk based base learner. Non-overlapped bag creation generates training subsets to handle data imbalance and the risk based Naïve Bayes eliminates the issues arising due to noise. Experiments were conducted and comparisons were performed with existing state-of-the-art fraud detection models, which indicates that NRBE exhibits improved performances of 5% in terms of BCR and BER, 50% in terms of Recall and 2X to 2.5X times reduced cost.