Stacking Model for Financial Fraud Detection with Synthetic Data
Zichuan Fu · Atlantis Highlights in Intelligent Systems/Atlantis highlights in intelligent systems · 2022
With the fast pace development of the Internet nowadays, financial frauds have also emerged continuously, which has seriously affected the development of the financial sector.Due to the lack of data in the financial field and the loose structure of transaction information, financial fraud detection remains a significant challenge.Based on the traditional machine learning model, this paper combines the three basic logistic regression models, support vector machine and random forest, and designs a two-layer stacking prediction model to detect financial transaction fraud.For unbalanced samples, this article uses up-sampling, under-sampling, and fusion methods to test and help search for optimal parameters through GridSearchcv.The final experiment shows that the Stacking model has a 97% recall rate and 87% accuracy for fraud samples on synthetic financial datasets.It can quickly detect most fraud samples while keeping false positives within a reasonable range.The model designed in this paper enriches the research of model fusion in financial fraud detection.