Non-Invasive Feature Selection for Intrusion Detection Systems in the Internet of Things

Illegal access to the Internet of things (IoT) and computer networks results in severe security damage, which must be protected by intrusion detection systems (IDS). In this paper, a novel intrusion detection method is developed for the computer networks and IoT, consisting of the hybrid algorithm of convolutional neural network (CNN) and bagging (BG) classifier. The optimal feature subset is carefully selected from the original feature space by its identifications or variable subset. The principal component analysis (PCA) transforms the original feature space into principal component space, where the variable subset is addressed by the sequential forward feature selection (SFFS) algorithm in combination with the random forest as the fitness function on the training set. The cross-validation procedure is used to estimate the intrusion detection performance of the proposed algorithm using the optimal feature subset on the testing set. The performance results show that the PCA and SFFS algorithms effectively improve selected feature subsets’ quality. The hybrid method of the CNN and BG classifier is well-fitted for applying the IDSs in practical IoT environments.