A new intrusion detection system for detecting ddos attacks using optimization and deep learning techniques

Abstract

Technological evolution is one among the cause of increase in various vulnerable cyber-attacks. In particular, involvement of Distributed Denial of Service (DDoS) attacks in various networking applications degrades the performance of the network servers by flooding the unwanted data to deny the services of the legitimate client. The proposed IDS have three works that are integrated as a single system. First work proposes a new Hyper Parameter Tuned and Recursive Feature Elimination (RFE) Aware Data Pre-processing Algorithm (HRDPA) for selecting the optimal features that is useful for making effective decision on network dataset. In addition, a new Deep Grid Network produces 6 best models for the effective prediction of existence of DDoS or Non-DDoS through the ensemble approach. Second work proposes a new feature selection technique called Split Filter Feature Selection and Spotted Hyena Optimization Based Feature Optimization Method (SFSH-FOM) for selecting the useful features that are used to enhance the performance of the classifier in terms of accuracy. In this work, a new cross layer feature fusion technique is also developed with the incorporation of FT-CNN and LSTM to improve the performance of the classifier. Third work proposes a new intrusion detection system to predict and detect the DDoS attacks effectively. In this work, a new Bayesian Fuzzy Rough Set theory-based Feature Selection Algorithm (BFRS-FSA) is developed to select the most contributed features which are used to improve the prediction accuracy. In addition, a new Gradient Boost Algorithm and CNN aware Hybrid Classifier (GCHC) is also proposed to predict and detect the DDoS attacks. This research work is evaluated by conducting experiments using KDD, CICIDS2017 and 2019 datasets and performance metrics such as precision, recall, F1-score and accuracy. Finally, this research proved better detection accuracy with less execution time and low false alarm rate. newline