An Improved XGBoost Model Based on Spark for Credit Card Fraud Prediction

摘要

Credit card fraud causes huge economic losses for many financial institutions. Given the imbalance of dataset and the huge amount of data in the field of credit card fraud, an improved XGBoost model based on Spark is proposed. In this project, the Smote algorithm was used to to balance the training set. And the XGBoost classifier based on Spark was used as the fraud detection mechanism. Finally, the test sets were classified in parallel. In the model comparison experiment, the model proposed in this project is compared with logistic regression model, decision tree model, random forest model, and original XGBoost model. The experimental results show that in the three metrics of Recall, Fl-Score, and AUC, the model proposed in this project is the best, which is 9.1%, 1.4%, and 1.2% ahead of the model ranked second respectively. In the speedup experiment, the speedup on the dataset of 70,000, 140,000, and 280,000 samples are 2.06, 3.28, and 3.75 respectively. The experimental results of these two parts show that the proposed model can accurately and efficiently predict credit card fraud and has a good practical effect.