Publication

Abstract : The rapid expansion of Internet of Things (IoT) devices across sectors such as smart cities, healthcare, and industrial automation has introduced major difficulties in maintaining network security and managing connected systems. Traditional IoT identification methods typically depend on standalone machine learning models, which often fail to cope with data imbalance and lack adaptability to diverse device types. To overcome these issues, this study introduces an ensemble learning-based framework that integrates Random Forest, Multi-Layer Perceptron (MLP), XGBoost, and AdaBoost using a soft voting mechanism. Using the Kaggle IoT Device Identification dataset—covering nine categories of devices and their flow-level network features—the proposed method applies dataset balancing and tree-based feature selection. Experimental evaluation shows that the ensemble model achieves 94% accuracy, 92% precision, 91% recall, and 92% F1-score, outperforming individual classifiers and other ensemble baselines. Confusion matrix results further validate its consistent accuracy across all device classes. This framework is optimized for deployment on IoT gateways and edge nodes, offering an efficient and scalable solution for real-time device recognition in dynamic network conditions.