HYBRID DEEP LEARNING AND ENSEMBLE INTELLIGENCE FOR ROBUST ZERO-DAY INTRUSION DETECTION IN IOT AND LOGISTICS CYBER SYSTEMS

Abstract

The rapid proliferation of Internet of Things (IoT) devices and the increasing digitalization of logistics and supply chain infrastructures have significantly expanded the cyberattack surface, making modern networks highly vulnerable to sophisticated and zero-day attacks. Traditional signature-based Intrusion Detection Systems (IDSs) often struggle to identify previously unseen threats and complex attack patterns in dynamic environments. This study proposes a comprehensive hybrid IDS framework that integrates deep learning, anomaly detection, ensemble learning, and federated learning techniques for detecting cyberattacks in both IoT and logistics network environments. The proposed architecture incorporates Transformer Autoencoders, Graph Attention Networks, Variational Autoencoders, Deep Autoencoder–Isolation Forest, CNN–LSTM, Federated Transformer IDS, XGBoost, TabNet, Random Forest, Extra Trees, and Hybrid Ensemble models. The framework is evaluated using the N-BaIoT IoT botnet dataset and a Zero-Day Logistics Network dataset. Experimental results demonstrate superior detection capability, where XGBoost achieved 99.997% accuracy and a ROC-AUC of 1.0000 on the N-BaIoT dataset, while ensemble-based models achieved perfect classification performance on the logistics dataset. The findings confirm that combining deep representation learning with ensemble intelligence enhances scalability, robustness, privacy preservation, and zero-day attack detection effectiveness in heterogeneous cybersecurity environments..