Search

In the latest years, with the integration of not only traditional networking devices but also a wide variety of Internet of Things (IoT) devices, connected devices retrieval gain much importance. This retrieval can be carried out through the use of various information research tools such as search engines. The devices type represents one of the most important attributes of connected devices retrieval, which helps effectively to give accurate results during the retrieval process. Efficient and sample connected devices classification and type identification represent a highly difficult task due to the wide variety of existing and evolving devices type. Identifying appropriate characteristics of connected devices may help simplify the classification process, which requires the integration of feature selection methods. For this, we analyze in this work the effect of feature selection to achieve a high performance and high accuracy of connected devices classification. To this end, we use different feature selection methods and evaluate these methods by applying a set of machine learning models. To extract the most representative features of our datasets we employed univariate feature selection, Recursive Feature Elimination (RFE), and Treebased feature selection (Random Forest). XGBoost, Decision Tree, and Random Forest are applied for performance evaluation based on the extracted feature. The evaluation results show that the selection of important features helps to improve the accuracy of connected devices classification using machine learning classifiers. [ABSTRACT FROM AUTHOR]