Improvement in credit card fraud detection using ensemble classification technique and user data

Financial fraud is a serious problem in banking system. Credit card fraud is growing with increasing Internet usage, as it becomes very simple to collect user data and do fraud transaction. Fortunately, all records including fraud and legit transactions are present in the financial record. Improved data mining techniques are now capable to find solutions for such outlier detections. Financial data is freely available in many sources, but this data has some challenges like,۱) the profile of legit and fraudulent behavior changes constantly, ۲) there is a class imbalance problem in dataset, because less than ۳% transaction are fraud, ۳) transaction verification latency is also one more problem. All this data issues are handled using pre-processing techniques like cleaning and reduction. Main aim of this research is to find out, output attribute "is Fraud", with better time complexity. To this end, K-means, Random Forest and J۴۸ algorithm is used, and its accuracy rates are compared to find best fit pre-processing and machine learning algorithm. It is observed that accuracy rate of Random Forest is ۹۳.۸% when both global and local dataset is used.

Keywords:

Credit card transaction , Global dataset , User dataset , J۴۸ , K-means clustering , Random forest , Ensemble method