The Influence of Device Type Aggregation on the Classification of Smart Home Devices Using Machine Learning Algorithms

The rise of the Internet of Things (IoT) has led to many different novel devices being introduced in smart homes. Automatic device classification techniques have therefore become necessary for effective network management and ensuring performance and security. Previous research in this area focused on datasets from single testbeds and applying Machine Learning (ML) algorithms to achieve device classification accuracy. The reported accuracy varies with the specifics of the testbeds and the datasets used, even if the same ML algorithms are being used. In our study, we investigated how much device type aggregation influences those results. We applied ML algorithms to the UNSW HomeNet dataset to analyze how classification accuracy varies with different device type labeling, providing insights into how device diversity affects classification performance. In particular, we investigated the influence of how device types are aggregated into classes. Our findings indicate that, while the accuracy fluctuates with an increased number of dataset classes, it stabilizes after accommodating a certain number of classes and devices. In addition, the study underscores the importance of the composition of the dataset, particularly the diversity of device types and manufacturers, in influencing the accuracy of ML algorithms.