Automated diagnosis of attacks in internet of things using machine learning and frequency distribution techniques

Abstract

The internet of things (IoT) is the interconnection of things around us to make our daily process more efficient by providing more comfort and productivity. However, these connections also reveal a lot of sensitive data. Therefore, thinking about the methods of information security and coding are important as the security approaches that rely heavily on coding are not a strong match for these restricted devices. Consequently, this research aims to contribute to filling this gap, which adopts machine learning techniques to enhance network-level security in the low-power devices that use the lightweight MQTT protocol for their work. This study used a set of tools and, through various techniques, trained the proposed system ranging from ensemble methods to deep learning models. The system has come to know what type of attack has occurred, which helps protect IoT devices. The log loss of the ensemble methods is 0.44, and the accuracy of multi-class classification is 98.72% after converting the table data into an image set. The work also uses a convolution neural network, which has a log loss of 0.019 and an accuracy of 99.3%. It also aims to implement these functions in IDS.