Filtering the Credit Card Fraud Detection Dataset for enhancing the Classification Performance

Abstract

Due to flexibility of payments the utilization and acceptance of credit cards is growing day by day. Additionally, the government is also forcing to transect through the online channels. This will help to improve transparency in payment systems. But the cases of financial fraud are also increasing. There available systems for credit card fraud detection are suffering to deal with the highly noisy data. Therefore we proposed a dataset quality enhancement technique. The proposed technique utilizes chi-square test, missing value handling, MTD technique for dimensionality reduction and regression technique for outlier detection. After enhancing the quality of data we utilize the refined data with the two machine learning algorithms namely Xgboost and Convolutional Neural Network (CNN). The experiments are carried out and performance is measured in terms of accuracy and training time. The comparative results demonstrate the Xgboost and CNN both are providing accuracy up 99% but the time utilization of CNN model is higher as compared to XgBoost.