Imbalanced Classification for Botnet Detection in Internet of Things

Abstract

The last decade has witnessed the widespread adoption of Internet of Things (IoT) in all spheres of human life be it manufacturing, logistics, medical, wearable devices, automobiles, smart grids, connected homes and industrial automation. Even though IoT creates a better experience for users and improved productivity and cost savings for industries, they are an attractive avenue for cyber-attacks. One of the most significant among attacks targeting IoT devices is botnet attacks. Botnet attacks leveraging IoT devices are found to be very damaging and recent cyber-attacks underscore this terrifying fact. In this paper, machine learning algorithms like decision tree, random forest, Naïve Bayes and logistic regression are used for detecting the botnet traffic. We have used BoT–IoT dataset published by UNSW Canberra. This dataset is highly imbalanced with very few non-attack samples. The research work has used synthetic minority over-sampling (SMOTE) with support vector machine (SVM) for addressing the problem of imbalanced classes. We have demonstrated the superior performance of SMOTE with SVM over Borerline1 SMOTE method and proposed technique out wins the usage of machine learning algorithms without any imbalanced class treatment. The proposed technique has shown significant improvement inaccuracy, precision, recall, F1-score, support and ROC-AUC score for detecting botnet samples.KeywordsIoTInternet of ThingsBotnetImbalancedMachine learningClassificationSecurity