IoT-based Network Attacks Discovery with Combined Classifiers

Abstract

In this paper following the recent trends in IoT-based network attacks discovery and advancing further our previous research, in which we optimize and test single neural network, support vector machine and random forest classifiers for both the detection and recognition of multiple DDoS attacks, we propose results from newly developed combined classifiers. The first of them employs only a neural network and a random forest classifier, while the second use additionally a support vector machine. Both are implemented in two modifications – as detectors of malicious vs. normal traffic, and as classifiers of 10 types of attacks vs. non-attack samples. High classification accuracy is being obtained over the popular Bot-IoT dataset and it prove higher than that of the single classifiers. At the same time, it is also higher than other solutions, proposed in the practice.