Effective Intrusion Detection through Hybrid CNN-LSTM and Grey Wolf Optimization for Feature Selection in Complex Network Environments

Abstract

Security and threat identification have always been critical in the interconnected world we live in today, and with emerging and cutting-edge threats, they still remain vital. Consequently, this paper proposes an intrusion detection solution whereby GWO for feature selection to optimize the detection model and a CNN–LSTM for attack detection. This paper describes the potential threat attributed to high dimensionality and non-linearity of network traffic data and aims to improve the overall detection rate and accuracy of different types of cyber-attacks. The GWO algorithm efficiently addresses the selection of relevant features from the data, hence demystifying the data while enhancing the model’s impact. CNN-LSTM architecture adapted from convolution neural networks and long short-term memory nets which performed the best in feature recognizing on networks traffic, allows the detection model to get understanding about complicated patterns in traffic. The described approach is compared with common methods, including Logistic Regression, Gaussian Naive Bayes, Decision Tree, and Random Forest and outperforms each of those regarding precision, recall and F1-score. Evaluation of the results obtained indicates that the proposed system correctly identifies various types of attacks with high precision, and very sensitive to false positive, and scenes the DoS/DDoS, PortScan, BruteForce, Web Attacks, and Botnet attacks correctly and distinctly. This paper shows the efficiency of integrating the state-of-art optimization methods with DL-based models for the development of the sound and highly scalable intrusion detection framework. The proposed method provides a much higher level of security enhancement in the proposed network against the emerging threat of cyber-attacks on existing and new complex networks.