Using Machine Learning to Build a Classification Model for IoT Networks to Detect Attack Signatures

Abstract

Internet of things (IoT) has led to several security threats and challenges within society. Regardless of the benefits that it has brought with it to the society, IoT could compromise the security and privacy of individuals and companies at various levels. Denial of Service (DoS) and Distributed DoS (DDoS) attacks, among others, are the most common attack types that face the IoT networks. To counter such attacks, companies should implement an efficient classification/detection model, which is not an easy task. This paper proposes a classification model to examine the effectiveness of several machine-learning algorithms, namely, Random Forest (RF), k-Nearest Neighbors (KNN), and Naïve Bayes. The machine learning algorithms are used to detect attacks on the UNSW-NB15 benchmark dataset. The UNSW-NB15 contains normal network traffic and malicious traffic instants. The experimental results reveal that RF and KNN classifiers give the best performance with an accuracy of 100% (without noise injection) and 99% (with 10% noise filtering), while the Naïve Bayes classifier gives the worst performance with an accuracy of 95.35% and 82.77 without noise and with 10% noise, respectively. Other evaluation matrices, such as precision and recall, also show the effectiveness of RF and KNN classifiers over Naïve Bayes.