Intrusion detection for IoT based on a hybrid shuffled shepherd optimization algorithm

The use of current data management technologies for the Internet of Things (IoT) is not straightforward due to interface heterogeneity, highly complex and potentially unprotected conditions, and the huge scale of the network. A vast number of devices and sensors around the world are connected to wireless networks, making the data transmitted to and from them vulnerable to attack. In addition, data sent through such devices can be unrelated, duplicated or inaccurate, making it impossible to execute the necessary tasks. Therefore, the transmitted data must be screened and chosen to fit the issue being handled in order to maintain the best possible degree of protection and thereby keep devices free from any data that could pose a threat. Reducing the size of the data often contributes to an increase in the level of communication between different networks, which in turn leads to a boost in the overall performance of the wireless network infrastructure. The process of obtaining only the relevant features from all of the data in a dataset for use in a particular task is known as feature selection (FS). In this paper, a new FS wrapper model is proposed that uses a shuffled shepherd optimization (SSO) algorithm as a search mechanism for the problem space and a K-nearest neighbor classifier to solve IoT problems with FS. Shuffled shepherd optimization was applied in a basic form as well as in two hybrid models with simulated annealing (SA) named SSO-SA1 and SSO-SA2. In order to test the performance of the three proposed models, they were applied to nine well-known IoT datasets. All the obtained results indicated that the two hybrids SSO-SA1 and SSO-SA2 enhanced the search capability of the original SSO algorithm. In comparison experiments, the SSO-SA2 outperformed three FS wrapper approaches, namely, EPC, MOPSO and MOPSO-Lévy. The SSO-SA2 surpassed the other methods in six out of the nine IoT benchmark datasets with an accuracy rate of 0.987. As for number of selected features, the SSO-SA2 surpassed the other methods in five out of the nine datasets with a selection size rate of 39.11 features. Furthermore, the wrapper-based SSO-SA2 surpassed four filter methods, namely, ReliefF, correlation, information gain and symmetrical.