Entropy-Based Feature Selection using Extra Tree Classifier for IoT Security

Abstract

      The Internet of Things (IoT) is a network of devices used for interconnection and data transfer. There is a dramatic increase in IoT attacks due to the lack of security mechanisms. The security mechanisms can be enhanced through the analysis and classification of these attacks. The multi-class classification of IoT botnet attacks (IBA) applied here uses a high-dimensional data set. The high-dimensional data set is a challenge in the classification process due to the requirements of a high number of computational resources. Dimensionality reduction (DR) discards irrelevant information while retaining the imperative bits from this high-dimensional data set. The DR technique proposed here is a classifier-based feature selection using an extra tree classifier (EXT). The entropy values of features are used for the construction of trees in EXT, which is to build a lower-dimensional space. Linear discriminant analysis (LDA), K-nearest neighbor classifier (KNN), decision tree classifier (DTC), and random forest classifier (RFC) empirically evaluate the proposed feature selection mechanism. EXT is compared with other DR techniques like RFC and principal component analysis (PCA). The performance metrics of the classifiers are used to evaluate the proposed work.