Deep Learning-Based Network Traffic Analysis For Intrusion Detection In Cyber Security Systems

Abstract

The development of internet-based applications has led to a more complex cyberspace, a major challenge to the cybersecurity experts in the development of intelligent real-time cybersecurity tools. In this paper, we propose a very effective intrusion detection model based on a hybrid GRU + BiLSTM on the BoT-IoT dataset. The suggested method combines data preprocessing, feature selection through Chi-square method, normalization, and data balancing through SMOTE to optimize model performance. The combination of GRU and BiLSTM allows for a model that detects malicious actions at extremely high rates by successfully capturing bidirectional and temporal patterns in network data. Experimental findings show high performance, with 99.1% precision, 99.3% F1-score, and 99.7% ROC-AUC, which show the good capability of classification and high robustness. The training and validation trends also indicate that the model has good generalization with little overfitting. The proposed approach is more effective than traditional and deep learning methods compared to the current models. In general, the research establishes that hybrid deep learning models are well fit to detect intrusion in IoT settings, and they offer effective and cost-effective security measures.