Enhancing Cybersecurity in the Internet of Things through Machine and Deep Learning : A Novel Approach to Threat Detection

The Internet of Things (IoT) is growing rapidly each day. By the increase of the number of internet-connected devices, security threats are getting more severe as well. Different machine learning models have been trained on malicious network traffic to detect security risks. In this study, we will test five machine learning techniques by manipulating training data to simulate a novel malware attack. Furthermore, we evaluate whether the trained models retain the ability to detect instances of excluded and therefore unobserved attacks in the training data. The Recurrent Neural Network performed best, with a weighted accuracy of 0.8173, followed by a Multilayer Perceptron Network with a weighted accuracy of 0.7828. This enhances our understanding of which machine learning techniques are most capable of detecting currently unknown types of attack and will improve our capabilities for future security.