Hybrid Ensemble Learning with Explainable AI for Anomaly Detection in Network Traffic
Anurag Gupta, Ajay Kumar Phulre, Adarsh M. Patel, Rizwan Ur Rahman · 2024
The increasing intricacy and quantity of network data pose significant obstacles for anomaly detection systems. Detecting and responding appropriately to novel and changing dangers is a challenge for conventional approaches. This research presents a novel method for explainable AI (XAI)-integrated hybrid ensemble learning for improving real-time anomaly detection in network data. The proposed model combines many machines learning techniques, including Random Forest, deep learning (CNN, Bi-LSTM) and boosting, to provide a robust ensemble capable of identifying a wide range of irregularities. Furthermore, by employing XAI techniques, the model's decisions are rendered in a clear and understandable way, encouraging confidence and comprehension among cybersecurity specialists. The proposed technique is tested on two widespread network traffic datasets: CICIDS 2017 and UNSW-NB15. Experimental results demonstrate that our model achieves high accuracy rates of 99.98% and 97.71% on these datasets, respectively, with False Positive Rates (FPR) as low as 0.02% and 0.01%. Based on experimental out-comes, our model provides a dependable solution for network anomaly detection, outperforming current state-of-the-art methods in rapports of accuracy, precision, recall, and F1-score.