A Fuzzy Intrusion Detection System for Identifying Cyber-Attacks on IoT Networks

The Internet of Things (IoT) is increasingly present in our daily activities, connecting the most varied types of physical devices present around us to the internet. IoT is the basis for smart cities, e-health, precision agriculture, among others. With this growth, the number of cyber-attacks against these types of devices and services has also increased. Each type of attack has its specific characteristics that allow its identification and prevention through machine learning techniques. However, classic machine learning techniques may have their performance compromised due to the non-stationary characteristics of these environments, together with the search for different types of vulnerabilities by attackers, attacks can suffer different types of mutations, in addition to the great possibility of new types of attacks arising over time. In this article, we propose an algorithm called Fuzzy Intrusion Detection System for IoT Networks (FROST) to identify cyber-attacks on IoT networks. FROST uses the concepts of fuzzy set theory to make the learning task more flexible, seeking to improve the performance in the classification of inaccurate data. In addition, FROST has a mechanism for identifying new types of intrusion online, during the classification of new instances. To evaluate our approach, we used the UNSW-NB15 data set and compared our method with another approach, very consolidated in the literature, which performs the same type of task. The results showed that FROST has a good performance in the classification of different types of attacks and that the fuzzy technique used helped to reduce errors and identify anomalies.