3RD INTERNATIONAL CONGRESS ON TECHNOLOGY - ENGINEERING & SCIENCE - Kuala Lumpur - Malaysia (2017-02-09)

Ensemble Of Classifiers For Credit Card Fraud Detection

Fraud detection is a crucial problem that has been facing the e-commerce industry for decades. Financial institutions throughout the world lose billions due to credit card fraud, which necessitate the use of credit card fraud detection system. Several studies have proposed models for fraud detection based on machine learning techniques like support vector machines, however, the accuracy of the model is crucial. Recently, in the area of machine learning, the concept of ensemble classifiers was introduced as a way for improving the classification accuracy by combining predictions from different classifiers. The paper proposes three ensemble models for credit card fraud detection which are: Bagging, boosting and stacking. Bagging and boosting models are based on K-nearest neighbors while Stacking model is composed of three base classifiers: Naïve Bayes, k-nearest neighbors and Decision trees and utilized the majority voting as a combination method. The performances of the three ensemble based models were compared to each other and to the performances of four fraud detection models developed using individual classifiers (Support vector machine, K-nearest neighbours, Decision Trees, Naïve Bayes). A real life anonymized data set of transactions (“UCSD-FICO Data Mining Contest 2009”) was utilized in developing and testing the models. Four popular metrics were used in evaluating the performances of the classifiers which are True positive rate (TPR), False Positive Rate (FPR), Balanced Classification Rate (BCR) and Matthews Correlation Coefficient (MCC). It was found that Stacking based model outperformed the rest of the models.
Marwan Fahmi, Abeer Hamdy, Khaled Nagati