Distributed Denial of Service attacks (DDoS) are one of the critical attacks in the network security field. This attack distracts and stops the service functioning of several online applications. Analysing cyber-attacks and anomalies is now becoming a highly severe issue in the cybersecurity domain. Hence, it has become complex to analyse these attacks and protect online services from them. Therefore, several deep learning techniques have also been incorporated for the earlier identification of DDoS attacks. However, the conventional model is also a tedious process and increases network complexity if the network size grows. Thus, it is computationally intensive and not well-suited for higher traffic volumes. Also, identifying the deviations from a large number of network patterns is highly scalable and generates false positives in dynamic network conditions. Therefore, it is necessary to control various problems that are identified in the existing multiclass DDoS attack analysis method with Machine Learning (ML) networks. In this work, at first, the necessary data for a multiclass DDoS attack classification system is assembled from an appropriate dataset. The whole dataset is split into training (75%) and testing (25%). In the APA-DDoS dataset, the total samples contain 798. Among them, the 598 and 200 samples are given to the training and testing process. In the CIC-DDoS2019 dataset, the total samples is 623. Here, the 167 and 456 samples are given to the training and testing process. In NSL-KDD, the total samples is 973. Among them, the 480 and 493 samples are given to the training and testing process. Hence, the collected data are further transmitted to the pre-processing stage. In this stage, the data cleaning and data normalisation methods are computed and processed from the collected data. Further, the pre-processed data is given to the multi-objective-aimed optimal feature selection stage, where the relief score, entropy and Chi-squared (Formula: see text Statistics are assumed. Additionally, the necessary features are chosen from the pre-processed data utilising the Position Revised Dung Beetle Optimiser (PRDO). Further, the tuned features are fed as an input to the Ensemble Deep Learning (EnDL) for classifying the multiclass DDoS attacks. The proposed ensemble learning network consists of various ML methods such as Capsule Network (CapsNet), Random Forest (RF), Multi-Layer Perceptron (MLP), Ridge Classifier (RC) and Support Vector Machine (SVM). Finally, multiclass DDoS attack classification results are obtained from the ensemble learning method. Moreover, different findings are computed in this proposed network to detect the efficiency over the traditional methods.
Prasanna et al. (Sat,) studied this question.