Explainable Hybrid Deep Learning Model for Cybersecurity
Ghazia Qaiser, Sivachandran Chandrasekaran, Jinchuan Zheng · 2025
In smart industries, cyber-attacks are the most consequential threats that can harm networks. Preventing cyberattacks from interrupting services is vital and challenging. Recently, deep-learning models have offered dynamic feature extraction and intelligent detection of zero-day attacks in complex architectures. This study proposes a novel hybrid deep learning model that combines the strengths of two of the most robust algorithms, MLP and BiLSTM, which are famous for detecting cyber-attacks. The study also proposed a vigorous and scalable hybrid deep learning pipeline designed for cybersecurity applications. The proposed model leverages behavioral analysis to detect malicious attacks. The proposed model was evaluated on the UNSW-NB15 and WUSTL-IIoT-2021 datasets and provided 99.70% and 98.75% detection rates, respectively. Model's interpretability is more enhanced using SHAP (SHapley Additive exPlanations) analysis, which provides feature-level explanations for predictions, offering more transparency in decision-making. Moreover, SHAP illustrates the impact of features on the model's prediction, which can help cyber security applications explore more and prevent zero-day attacks. The explainability of the model can offer behavioral profiling and enhance the state-of-the-art in anomaly detection for cybersecurity.