Automobile Finance Credit Fraud Risk Early Warning System based on Louvain Algorithm and XGBoost Model

With the rapid growth of the automobile finance credit market, fraud risks—particularly organized and covert activities—pose significant challenges to traditional single-dimensional risk control methods. To address this, this study proposes a multi-dimensional fraud risk warning framework integrating complex network analysis and machine learning. First, the Louvain algorithm is employed to detect tightly-knit fraud communities in social networks by optimizing modularity. Concurrently, an XGBoost model analyzes customer profiles, behavioral sequences, and network topology features to predict individual fraud risks. A dynamic weight fusion mechanism is then designed to adaptively combine outputs from both methods, prioritizing high-confidence predictions from either model. Experimental results demonstrate superior performance, achieving 92% accuracy, 95% recall, and a 0.97 AUC, outperforming traditional approaches. The integration of community detection and adaptive fusion effectively identifies cross-platform fraud patterns, offering a novel technical pathway for enhancing financial risk control through heterogeneous data synergy and decision optimization.