Decision Analysis and Prediction Based on Credit Card Fraud Data

With the common use of credit cards in today's transactions, the related fraudulent behavior inevitably occurs and causes considerable loss of money. To solve this problem, our work used a dataset that contains legal credit card transactions as well as fraud transactions to find an effective solution. In this paper, through processing and analyzing the transaction data, the data was discovered to be unbalanced, so stratified sampling and oversampling were performed to achieve a more reliable analysis of the unbalanced dataset. Meanwhile, due to the randomness of sampling, the cross-validated were used for the final model selection. Further, we utilized five algorithms to build models which contains statistical machine learning model and deep learning model. To obtain optimal performance of the models, hyperparameter tuning was performed for the five classifiers. Finally, the results indicate that the optimal model was XGBoost, and its performance can be verified in a real-life scenario in the future.