Risk based bagged ensemble (RBE) for credit card fraud detection

Credit card frauds costs consumers several billions of dollars annually. Even with several systems in place accurate fraud detections remains unsolved, due to several intrinsic issues contained in transaction data. This paper analyzes the intrinsic nature of data and proposes a risk based ensemble model RBE as a solution to handle the intrinsic issues contained in data and also to provide effective results. The conventional bagging model is extended and novel enhancements have been incorporated in terms of an effective base learner and a cost sensitive combiner. Bagging models are highly efficient in handling imbalanced data, while incorporation of risk based Naïve Bayes handles the implicit noise contained in transaction data. Cost sensitive combiner replaces the conventional voting combiner to produce results that exhibits high performances and low cost. Comparisons with state-of-the-art models indicates the high performance levels of the proposed RBE model.