Research on Network Threat Hunting System Based on Multi Scale LightGBM Ensemble Learning

With the increasing complexity and concealment of network attacks, traditional single-scale threat detection methods have made it difficult to meet the needs of modern network security. This study proposes a network threat-hunting system based on multi-scale LightGBM ensemble learning, aiming to improve the accuracy and efficiency of threat detection by fusing network data at different time scales and spatial scales. Firstly, the system extracts multi-scale features from network data, including real-time traffic, historical behaviour and topology, and then uses the LightGBM algorithm for ensemble learning. The experimental results show that the threat detection accuracy of the multi-scale feature fusion model is improved by 15.3%, which is significantly better than the single-scale model. At the same time, the LightGBM ensemble learning model performs well in detection efficiency, and the average detection time is shortened by 20.7%. The generalization ability of the system in different network environments has also been verified, and the average threat detection recall rate reaches 92.1%. These results show that the multi-scale LightGBM ensemble learning system performs well in terms of accuracy, efficiency, and generalization ability, providing a new solution for cyber threat detection.