Machine Learning for IoT Devices Security Reinforcement

As more lightweight objects connect to the Internet, the Internet of Things (IoT) is changing our linked environment. Thus, IoT intrusion detection research must be high-quality to provide solutions. Network intrusion datasets are essential for training and testing attack detection algorithms. This study describes, statistically analyzes, and machine learning evaluates the innovative ToN IoT dataset. Heterogeneity in IoT datasets is important and can affect detection performance. We demonstrate through a cross-validated experiment with five classifiers that industry-useful IoT network intrusion datasets require several data collection methods and a wide variety of monitored variables. Among our five models, RF emerged as the best one achieving an impressive accuracy of 0.991 with an Area Under the Curve of 0.9996. Moreover, its training time took only 0.985 seconds. Our study reviews an important amount of datasets focused on IoT security. We will see that standardizing feature descriptions and cyberattack classifications is necessary for the operational use of IoT datasets.