An Efficient CatBoost Classifier Approach to Detect Intrusions in MQTT Protocol for Internet of Things

Recent advancements in Internet of Things (IoT) infrastructures attribute a rise in undesirable issues specific to network security. As the number of IoT devices connected to the network rises daily, the network is more vulnerable to cyber-attacks. Hence, an intrusion detection system (IDS) is vital for detecting the type of cyber-attacks automatically in a time-bound manner. Moreover, the network often uses the MQTT protocol to deploy communication among IoT devices. This work proposes a CatBoost algorithm, a variant of machine learning (ML) algorithms, to classify the given attack into SlowITe, Malformed, Brute force, Flood, Dos, and Legimate. The algorithm is trained on a publicly available MQTT network dataset by creating a balancing dataset. Despite the significant disparity in the number of labeled records for each dataset class, the algorithm achieves state-of-the-art performance. The test result suggested that the algorithm can classify the type of attack with an accuracy of 94% within 78.45 s in the balanced dataset.