IoT Device Classification Using Hybrid Stacking Machine Learning Model on IoT Network Traffic Data

In the expanding domain of the Internet of Things (IoT), accurately classifying devices is essential for ensuring secure and efficient network management. As IoT networks increase in size and complexity, the challenge of effectively managing these devices becomes more pressing, particularly in defending against the risks posed by Shadow IoT devices unauthorized devices that connect to networks without formal approval. Despite some progress, previous research has largely focused on individual machine learning algorithms or simple ensemble models, with limited exploration of hybrid stacking models that combine two classifiers. This leaves a gap in understanding how different classifier combinations can balance accuracy, training time, and prediction time. This thesis addresses these gaps by using a hybrid stacking machine learning model that integrates two classifiers with the focus on reducing training and prediction times while maintaining or improving classification accuracy. The results show significant improvements in existing models. Combination of Random Forest (RF) and XGBoost (RFXGB) achieved a test accuracy of 72.4% and a training accuracy of 99.2%, with a training time of 4.17 seconds and a prediction time of 0.053 seconds, marking a 7.4% increase in test accuracy compared to the single XGBoost classifier. Similarly, the combination of Random Forest and Support Vector Machine (RFSVM) achieved a test accuracy of 71.2% and a training accuracy of 99.6%, with a training time of 2.196 seconds and a prediction time of 0.051 seconds, showing a 12.6% improvement in test accuracy over the single Support Vector Machine classifier.