An Ensemble-Based Method for DDoS Attack Detection in Internet of Things Network

This paper proposes an ensemble-based framework for Distributed Denial of Service (DDoS) attack detection in internet of thing (IoT) network. A combination of k-NN, MLP, and DNN classifiers are used to make an ensemble framework. Final predictions are determined by a weighted voting where weighted outputs of the best two classifiers are used. Experiments are executed on two recent IoT datasets: ToN-IoT and IoT23 datasets. To improve the classification accuracy, the datasets are subjected to a variety of pre-processing approaches and feature selection processes. The feature selection is handled through the combination of the Pearson correlation coefficient, entropy and mutual information to avoid redundant data and obtain improved feature sets. In comparison to other relevant research utilizing the same dataset, experimental results demonstrate that the ensemble method achieved more satisfactory results in terms of accuracy, precision, and the receiver operative characteristic (ROC) curve and provided a considerable improvement. The ensemble based model records a detection accuracy of 99.998% and ROC of 99.995% which shows that its ability to classify attack cases from benign is superb. The effect of feature selection methods on the performance of the ensemble model is also investigated and discussed.