Feature Generation by K-means for Convolutional Neural Network in Detecting IoT System Attacks

The research presents a new method to detect attacks on the Internet of Thing (IoT) system using the convolutional neural networks combined with the K-means clustering algorithm. Current studies on machine learning and deep learning also use independent features that have not yet taken advantage of the role of feature groups and the relationship between features. Therefore, in this paper, we use the K-mean algorithm to cluster feature groups, find good groups and generate new linked features from good feature sets to expand the original set of features. From the improved feature set, we use CNN to detect IoT system attacks. The method proposed in this paper is tested on three different data sets, achieving an average accuracy of 97.87% and improved 1.018% compared with the detection method using only CNN.