An Adaptive Hybrid Deep Learning Framework for Cybersecurity Threat Prediction and Anomaly Detection

Abstract

Threats have been complex and newly developed; consequently, cybersecurity is a major concern in the present connected environment because of their rapid evolution. The legacy rule-based intrusion detection systems can no longer detect complex or novel threats. We propose a new adaptive deep learning architecture including Convolutional Neural Networks (CNN), Bidirectional Long Short-Term Memory (BiLSTM), and Attention Mechanisms to improve threat-prediction power and anomaly detection in cybersecurity. The proposed model is subjected to the entity [\"dataset\",\"CICIDS2017\",\"cybersecurity intrusion detection dataset\"] dataset. The performance measures also demonstrate an enhancement in detection accuracy, decreasing false-positive rates, and robustness under various attack scenarios. Further confirmation of the efficiency of each single element of the proposed architecture is the introduction of ablation studies and security analyses.