Explainable Risk Aware Hybrid Intrusion Detection System Using Machine Learning and Deep Learning

The rapid growth of cyber threats has exposed the limitations of traditional intrusion detection systems, particularly their inability to detect complex attacks and provide interpretable results. This paper proposes an explainable and risk-aware hybrid intrusion detection system that integrates machine learning and deep learning techniques to enhance detection performance and transparency. The proposed framework combines Random Forest and Logistic Regression with a Long Short-Term Memory (LSTM) network to capture both statistical and sequential patterns in network traffic. Experiments conducted on benchmark datasets such as CIC-IDS2017 and NSL-KDD demonstrate that the hybrid model outperforms individual models in terms of accuracy, precision, recall, and F1-score, achieving detection accuracy of up to 94–97% while reducing false alarm rates. Furthermore, a risk scoring mechanism is introduced to classify detected intrusions into multiple severity levels, enabling prioritized threat response. To improve interpretability, SHAP-based explainable AI techniques are employed to provide both global and instance-level insights into model predictions. The results indicate that the proposed system enhances both performance and trust, making it suitable for practical cybersecurity applications.