Hybrid Deep Representation and Ensemble Learning for Secure and Adaptive IoT Device Classification

Abstract

The rapid expansion of the Internet of Things (IoT) has resulted in a massive network of interconnected devices spanning sectors such as healthcare, smart infrastructure, and industrial systems. Despite its advantages, this widespread connectivity exposes networks to critical security challenges, including unauthorized intrusions, data manipulation, and service disruptions. Traditional security mechanisms like Access Control Lists (ACLs), rule-based systems, and signature-driven detection approaches are becoming less effective due to their static configurations, reliance on frequent manual updates, and limited capability to detect evolving threats. To address these issues, this research introduces a smart IoT security model leveraging a Deep Autoencoder (DAE) for dual purposes: device classification and data validation. The DAE extracts compact and meaningful feature representations from IoT traffic, allowing efficient identification of normal and malicious device behaviour. The model is trained on labeled datasets to uncover complex traffic patterns and detect anomalies with improved accuracy. Additionally, an embedded authentication component verifies the integrity of device communications before granting network access, enhancing trust within the system. Experimental results show that the proposed framework achieves an accuracy of 97.8%, outperforming conventional methods such as KNearest Neighbors (KNN) and Logistic Regression Classifier (LRC). Furthermore, a hybrid approach named DAE-BFL (BFL), which integrates DAE with Random Forest Classifier (RFC) and LRC, provides enhanced classification robustness. The complete workflow includes preprocessing, training, validation, and real-time evaluation, ensuring scalability, adaptability, and effectiveness in securing modern IoT environments