Detection of DDoS Attacks in IoT Devices

Detection of DDoS attacks is one of the challenging tasks so that only legal users can use the proper resources. In this paper, the detection of attacks using various machine learning classifiers is discussed [1]. By utilising the three most pertinent feature selection techniques (chi-squared, RFE (recursive feature elimination), and reliefF), the most significant features were extracted from a publicly accessible NSL KDD dataset. Then, from the entire merged feature set, we selected the most relevant features by using a hybrid method. The dataset acquired using the hybrid technique underwent instance filtering to separate DDoS instances because all anomalous cases do not constitute DDoS instances, and the dataset was given the name train DDoS. This train DDoS dataset was uploaded into the weka tool and discretized using the discretise tool. Finally, the detection rates of the dataset were calculated by applying four machine learning classifiers (Naive Bayes, Decision Table, SVM (support vector machine) and Random Forest) and the detection rates were plotted and compared. The hybrid method produced the best detection rate of 99.97% and an average detection rate of 95.18% using the same set of classifiers as the train set, which had a best detection rate of 99.99% and an average detection rate of 86.41%. Thus, in comparison to other methods, the hybrid method has the best detection rate [2].