Dynamic ensemble based feature selection model for credit card fraud detection

Credit card fraud has increased significantly with the increased use of credit cards. So an efficient but robust system is needed to detect credit card fraud. Many machine learning algorithms have been proposed for this purpose. Among them, recently proposed dynamic ensemble classification using soft probability (DECSP) outstood other individuals as well as ensemble classifiers. However, the authors do not focus on feature selection/extraction, which can improve the model to a great extent. Hence, we propose a framework for credit card fraud detection based on DECSP. The framework presented incorporates DECSP with various feature selection/extraction techniques such as Particle Swarm Optimization (PSO), Genetic Algorithm (GA), Principal Component Analysis (PCA) and Recursive Feature Elimination (RFE). We experimented and compared the results generated based on accuracy, precision, f1-score, and AUC. DECSP with GA outperforms other techniques on all four performance metrics. Further, we validated the results based on statistical tests.