An efficient attack detection approach for software defined Internet of Things using Jaya optimisation based feature selection technique

Software defined Internet of Things (SD-IoT) has drawn several attacks because of its novelty. To counter such attacks, this paper presents a study on feature selection using Jaya Optimisation for making the lightweight intrusion detection system (IDS) for the data plane of SD-IoT. To check the effectiveness of the selected features, an ensemble of tree-based classifiers that uses a boosting approach called light gradient boosting machine (LGBM) is trained and tested with all features (AF) and then with selected features (SF) using 10-fold cross-validation. It was found that LGBM gave better performance when it was trained with SF. For performance evaluation, some well-known metrics have been used, namely recall, accuracy, false alarm rate (FAR), F1, precision, Cohen's Kappa coefficient (CKC), and prediction time. This trained model is deployed for attack detection in the OpenFlow-enabled devices of data plane of SD-IoT where it can detect the attacks in a distributed manner.