XGBoost Based Supply Chain Fraud Detection Model

Abstract

There are plenty of product data in today's world, and increasing companies pay more attention to using the data effectively. And the application of machine learning models to detect product Fraud has received widespread attention from researchers and enterprises, which could help companies adjust their plans to adapt to the market and increase revenue. In this paper, we hybridize the XGBoost algorithm and Random Forest, and then get an optimized model whose output is the average of the output of the XGBoost model and the Random Forest model. It shows that the Confusion matrixes for XGBoost and Random Forest, the quantity of TP and TF is large, which means the hybrid model performs well. Moreover, I evaluate the hybrid model on the DataCo smart supply chain datasets provided by the Kaggle competition. Experimental results show my method achieves superior performance over the other machine learning approaches. My model's F1 score is 0.49, 0.49, and 27.9 higher than the Logistic regression algorithm, the SVM, and the Gaussian Naive Bayes algorithm respectively.