CatBoost for Fraud Detection in Financial Transactions
Yeming Chen, Xinyuan Han
Abstract
Financial fraud is an ever growing menace with severe consequences in the financial industry. Machine learning plays an active role in the fraud detection in financial transactions. However, fraud detection is still a challenging problem due to two major reasons. First, either fraudulent or non-fraudulent behaviors change fast and constantly. Secondly, currently online transactions happen so fast, which require detection algorithms to be efficient and accurate. This paper introduces a machine learning method based on CatBoost for fraud detection. To improve detection accuracy, we apply feature engineering to generate highly important features and feed them into CatBoost for classification. Another key contribution of our work is using memory compression to speed up detection. The performance of our method is evaluated on a publicly IEEE-CIS Fraud dataset, provided by Kaggle competition platform. The experimental results demonstrate that our model based on CatBoost has obtained optimal accuracy of 0.983.