Insurance Fraud Prediction Model Based on eXtreme Gradient Boosting

As the volume of insurance transactions continues to increase, combating insurance fraud has become increasingly important. One of the key challenges is how to predict fraud based on high-dimensional data. In the current research landscape, deep learning methods that perform well often require large amounts of data, but their training and deployment also pose significant computational challenges. This paper, based on the Alibaba Tianchi dataset, explores the feasibility of constructing a low-cost prediction system through feature engineering and parameter tuning. Additionally, this paper verifies the robustness of the eXtreme Gradient Boosting (XGBoost) model in handling high-dimensional sparse features and imbalanced data.