An Ensemble Method for IoT Device Classification
Qi Shen, Zuoying Zeng · 2022
More and more Internet of Things (IoT) devices are allowed to connect to the networks conveniently, and the network providers need to identify which devices are safe. However, IoT devices have a fast update rate, and organizers such as network providers are not able to identify and classify the types of devices connected to the network well, which by establishing the corresponding rules promptly. Though there are a large number of self-supervised, semi-supervised, and unsupervised learning applications for classification tasks, supervised learning of traditional machine learning methods is still a very important part of the classification task. In particular, the sample sizes of newly emerging devices are usually small, and deep learning cannot take advantage of the ability that learn a large number of samples. This paper used traditional machine learning methods such as random forest, logistic regression, k-nearest neighbor, and Gaussian Bayes, as well as selecting the boosting methods that have previously demonstrated SOTA performance, such as XGBoost, LightBGM, and GradientBoost for training. Based on this, the performance was improved by combining classifiers using the stacking and voting model, and the optimal weight assignment in the voting methods was proposed.