Intelligent Cyber Attack Detection with Enhanced Feature Representation

The rapid escalation of cyber threats has exposed critical limitations in existing intrusion detection systems, particularly regarding real-time processing, detection of zero-day attacks, and handling imbalanced network traffic. This paper proposes a novel Hybrid CNN-LSTM-Attention (HCLA) framework that integrates convolutional neural networks for spatial feature extraction, long short-term memory networks for temporal dependency modeling, and multi-head attention for adaptive feature weighting. The proposed architecture addresses key research gaps by effectively capturing both local patterns and long-range dependencies while maintaining computational efficiency. Comprehensive experiments on the CICIDS۲۰۱۷ dataset demonstrate that HCLA achieves ۹۸.۷% accuracy, ۹۸.۵% precision, ۹۸.۳% recall, and ۹۸.۴% F۱-score, outperforming existing CNN, LSTM, GRU, and traditional machine learning baselines by ۲.۱% to ۵.۸% in F۱-score. Ablation studies confirm the contribution of each architectural component, with the attention mechanism providing ۲.۳% performance improvement. The framework exhibits inference latency of ۲.۳ms per sample, enabling real-time deployment in high-speed network environments.

کلیدواژه ها:

Cyber Attack Detection ، Deep Learning ، CNN-LSTM ، Attention Mechanism ، Intrusion Detection ، Network Security