A New Improved Method for Online Credit Anti-Fraud
Haiyan Kang, Hao Zhang · Automatic Control and Computer Sciences · 2022
Abstract With the rapid development of Internet finance, the demand for online credit anti-fraud is more and more urgent. The current research on credit anti-fraud is constantly advancing. In response to the imbalance of credit fraud, a new improved online credit anti-fraud method based on KCV-SMOTE (K-fold cross-validation and synthetic minority oversampling technique) and KFS (key feature scanning) is proposed in this paper. This method firstly performs KCV-SMOTE on the original training data to obtain a synthetic training set. Secondly, this method performs key feature scanning on the synthetic training set to obtain a set of sub-training sets. Then, this method trains the LR (logistic regression) classifier separately by using the sub-training sets and selects the intersection of the sub-training sets of the optimal classifier to train the optimal LR classifier. This method finally uses the optimal LR classifier to obtain the final prediction result. Experimental result shows that the method proposed in this paper increases the AUC value of the classification to 98.6221%.