Explainable Hybrid Tree-Based Learning for Robust IoT Device Classification and Behavioural Pattern Analysis

Abstract

The rapid growth of Internet of Things (IoT) technologies has led to the widespread deployment of connected devices across smart homes, industries, and urban infrastructures. The increasing volume and heterogeneity of IoT data demand intelligent analytical systems capable of understanding device behaviour and ensuring efficient management. Traditionally, IoT device classification relied on rulebased and statistical approaches, which struggled to handle high-dimensional data, resulting in low accuracy, poor scalability, and limited adaptability to dynamic environments. As IoT ecosystems evolved, the adoption of advanced Machine Learning (ML) techniques became essential to address these limitations. This research presents an interpretable and efficient analytical framework for accurate IoT device classification. The proposed system integrates data preprocessing, exploratory data analysis, and multiple Machine Learning (ML) models, including Gaussian Naive Bayes (GNB), Multinomial Naive Bayes (MNB), and Decision Tree (DT), along with a hybrid Denoising Autoencoder Tree (DAETree) model that combines Denoising Autoencoder (DAE) and Greedy Tree (GT). To address class imbalance, the Adaptive Synthetic Sampling (ADASYN) technique is incorporated, generating synthetic samples for minority classes and improving model generalization. The hybrid model enhances feature representation by capturing complex data patterns while maintaining interpretability through tree-based decision structures. Experimental results demonstrate significant performance improvements over traditional methods, with the DAE-Tree model achieving a highest accuracy of 1.0000 (100%), outperforming GNB, MNB, and DT models. The system supports both real-time and batch prediction, ensuring scalability and practical applicability. The significance of this study lies in its ability to combine interpretability, balanced learning, and high predictive performance, making it suitable for real-world IoT applications such as smart homes, industrial automation, and intelligent monitoring systems.