Zero-Day Cyberattack Detection Using Federated Deep Autoencoder and Explainable AI Framework

The rapid growth of cyber threats and zero-day attacks has exposed the limitations of traditional signature-based intrusion detection systems. To address this challenge, this paper proposes a Federated Deep Autoencoder with Explainable Artificial Intelligence (FDA-XAI) framework for privacy-preserving zero-day cyberattack detection. The proposed framework combines federated learning, deep autoencoder-based anomaly detection, and SHAP-based explainability to identify malicious activities without sharing sensitive organizational data. Multiple organizations collaboratively train a global intrusion detection model while maintaining data privacy. Experimental evaluation using the CICIDS2017 and UNSW-NB15 datasets demonstrated that the proposed FDA-XAI framework achieved 98.21% accuracy, 97.85% precision, 97.42% recall, 97.63% F1-score, and 0.991 ROC-AUC. Furthermore, the model achieved a 96.8% zero-day attack detection rate, outperforming conventional deep learning approaches. The results indicate that the integration of federated learning and explainable AI provides an effective, scalable, and privacy-preserving solution for next-generation cybersecurity systems