Machine learning-based solutions for securing IoT systems against multilayer attacks

IoT systems are prone to security attacks from several IoT layers as most of them possess limited resources and are unable to implement standard security protocols. This paper distinguishes multilayer IoT attacks from single-layer attacks and investigates their functioning. For developing a robust and efficient IDS (intrusion detection system), we have trained a few machine learning (ML) approaches such as NB, DT, and SVM using three standard sets of IoT datasets (Bot-IoT, ToN-IoT, Edge-IIoTset). Instead of using all features, the ML models are trained with similar features of multilayer IoT attacks to use optimal computational power and minimum number of features in the training dataset. The NB model achieves an accuracy of 57%–75%, while the DT model achieves an accuracy of 93%–100%. The outcome of the two ML models reveals that training with similar features possesses a higher accuracy level.