Secure coding for the Internet of Things (IoT)

Abstract

Key words: Secure, coding, IoT, Internet Of Things, IoT, Malware, Machine Learning, Deep Learning, Dataset, attack, UNSW-NB15, Network Intrusion Detection, NIDS, Hierarchical

Internet is not only about humans accessing the Internet through their mobile phones or laptops but it’s extended to a plenty of devices like refrigerators, air conditioners, cars, light bulbs...etc. Therefore we have IoT. Currently IoT is a very important scope of research since it’s connecting the whole world together. IoT has wide Economic, Industrial, Health benefits and many more. IoT devices are easy accessible and widely used, this caused many security challenges. One of the most challenging security problems in IoT is Network attacks like: worms, exploits, DoS ...etc. Therefore, a Network Intrusion Detection System is extremely important for a more secured IoT eco-system. From the most proven to be effective methods for malware detection recently is Machine learning. That’s why in this thesis we present a cascaded NIDS in IoT using machine learning algorithms. The main purpose behind this research is presenting a NIDS that gives a good accuracy with good complexity. It detects the normal/abnormal traffic and if the traffic is abnormal then it will identify the type of abnormal traffic. Cascading was prefered for less complexity and better accuracy. We used in this research of the most recent and comprehensive data set in the latest 5 years which is UNSW-NB15 data set which contains a lot of modern IoT attacks. The experiment performed showed that Random Forest is the best algorithm for either binary (with accuracy 99.6%) or multi- class classification (with accuracy 90%). Also we used feature reduction to reduce the UNSW-NB15 features from 47 to 15 features.