Credit Card Fraud Detection by Modelling Behaviour Pattern using Hybrid Ensemble Model

The fraud detection system in banking organisation relies on data-driven approach to identify the fraudulent transactions. In real time, detection of each and every fraudulent transaction becomes a challenging task as financial institutions need aggressive jobs running on the log data to perform a data mining task. This paper introduces a novel model for credit card fraud detection which combines ensemble learning techniques such as boosting and bagging. Our model incorporates the key characteristics of both the techniques by building a hybrid model of bagging and boosting ensemble classifiers. Experimentation on Brazilian bank data and UCSD-FICO data with our model shows sturdiness over the state-of-the-art ones in detecting the unseen fraudulent transactions because the problem of data imbalance was handled by a hybrid strategy. The proposed method outperformed by a margin of 43.35–68.53, 0.695–11.67, 43.34–68.52, 42.57–67.75, 3.5–13.06, 24.58–34.35%, respectively, in terms of true positive rate, false positive rate, true negative rate, false negative rate, detection rate, accuracy and area under the curve from the state-of-the-art-techniques, with a Matthews correlation co-efficient of 1.00. At the same time, the current approach gives an improvement in the range of 0.6–24.74, 0.8–24.80, 10–17.00% in terms of false positive rate, true negative rate and Matthews correlation co-efficient respectively from the state-of-the-art techniques with detection rate of 0.6650 and accuracy of 99.18%, respectively.