An Effective Deep Learning Model Design for Cyber Intrusion Prevention System

Objectives: The increasing frequency of cyber threats necessitates the advancement of Intrusion Prevention Systems (IPS). However, existing IPS models suffer from high false positive rates, inefficiencies in real-time detection, and suboptimal accuracy levels. Methods: This study presents a CNN-LSTM hybrid model optimized for real-time cyber intrusion detection. The CICIDS2018 dataset was utilized for training, incorporating feature selection, hyper-parameter tuning, and dropout-based regularization to improve efficiency and prevent over-fitting. Findings: The proposed system achieved an F1-score of 99.5%, significantly outperforming conventional methods. Additionally, the false positive rate was reduced by 18%, enhancing system reliability in cyber-security applications. Novelty: Unlike prior works, this study integrates optimized feature selection mechanisms with real-time sequence learning through CNN-LSTM, leading to higher detection accuracy, improved generalization, and reduced computational complexity. Keywords: Convolutional neural networks (CNNs), CICIDS2018, Deep Learning, Feature selection, Long Short­term Memory Networks (LSTMs)