A Novel Deep Learning Based Network Hybrid Intrusions Detecting System using Dense Layers, Bi LSTM with Multi-Head Attention, and XGBoost Classification

Abstract

As cyber threats have been evolving at a very fast pace, conventional Network Intrusion Detection Systems (NIDS) tend to lose high precision and have low false positive rates. In this paper, we introduce a hybrid method for intrusion detection based on networks by combining deep learning with machine learning methods. Our model utilizes Dense Layers and Bi-LSTM networks with Multi-Head Attention for extracting spatial-temporal features from network traffic flows. To improve classification accuracy, we use XGBoost as a decision-making layer. The Multi-Head Attention helps the model concentrate on key features in sequential data, while XGBoost maximizes classification accuracy with its fast gradient-boosting paradigm. The approach is tested with benchmark intrusion detection datasets and is found to perform better than the traditional approach with regard to accuracy, precision, recall, F1-score, and ROC-AUC. This hybrid solution provides a strong and scalable real-time intrusion detection solution that greatly enhances network security and threat mitigation. Experimental results show high detection accuracy and efficiency, performing better than existing models in classifying sophisticated attack types like DDoS and U2R. The solution is tested on CIC-IDS2017, CIC-DoS2017, and CSE-CIC-IDS2018 datasets, yielding significant improvements in precision, recall, and F1-score.