Enhancing Intrusion Detection with GANs for Class Balancing and Random Forest Classification

This study explores how to improve prediction accuracy in network intrusion detection by addressing class imbalance issues. With the rapid development of the Internet, network security threats are increasingly complex, which puts forward higher requirements for traditional intrusion detection systems. This research presents an integration of Generative Adversarial Networks (GANs) with Random Forest Classifiers, and conducts a comparative analysis of the performance of various models, including XGBoost, LightGBM, and H2O AutoML. The findings validate the superiority of the Random Forest model in addressing imbalanced datasets, achieving a prediction accuracy of up to 87.14%, which is approximately 10% higher than that of other models. In addition, the feature importance analysis within the Random Forest model shows that features such as ‘Dst_Host_Srv_Count’ and ‘Srv-Count’ contribute the most in prediction, mainly reflecting anomalies in network traffic patterns and connection frequencies. The research provides a practical solution for the field of intrusion detection, which helps to improve the overall protection capability of network systems. By understanding the importance of key features, network security strategies can optimize resource allocation and implement more targeted monitoring, thereby improving the ability to respond to complex attacks.