Improvement of attack detection performance on the internet of things with PSO-search and random forest

The presence of the internet of things allows various smart devices to be connected and interact with each other. Although IoT provides benefits in daily activities, however, with the presence of new technologies, IoT is vulnerable to new types of attacks. The massive IoT traffic results in a large number of traffic features and constructs complex network that makes intrusion detection systems (IDSs) require large resources to identify the type of attacks. On the other hand, most of the intrusion detection techniques are not feasible for IoT networks because they require more computing resources for attack detection, while IoT devices have limited computing resources and storage capacity. Thus, a lightweight IDS that has ability to identify new types of attacks is required. This research proposes a hybrid of Panigrahi and PSO-Search approaches to reduce the complexity of the network by eliminating the number of irrelevant features effectively and efficiently and combine with Random Forest optimization method to improve detection performance. The proposed IDS is validated with training and testing data, using hold-out, Stratified k-fold cross-validation, and percentage split test mode on CICIDS-2017 dataset MachineLearningCSV version. The dataset is chosen, as it represents real IoT network traffic data. Experimental results show that the performance improvement of the proposed hybrid IDS is very encouraging. The accuracy rate reaches 99.9 %, with an average Recall value of 1.000.