A Dynamic BPN-MLP Neural Network DDoS Detection Model Using Hybrid Swarm Intelligent Framework

Abstract

Background/Objectives: The most untreated and severe cyber security issue in cloud computing is DDoS attack, this is being under research to find novel findings with less complexity and better efficiency to detect and mitigate this issue. In this research article, Artificial Neural Network (ANN) algorithms like Backpropogation neural network (BPN) and Multilayer perceptron (MLP) are implemented and their performance on intrusion detection by utilizing NSL-KDD dataset is demonstrated. Methods: Initially, NSL-KDD benchmark dataset construction is carried out in the range of (0-1) using min-max normalization technique. Following this, hybrid Harris Hawks optimization particle swarm optimization (HHO-PSO) is employed to reduce the dataset size by selecting significant features that represents anomaly in network traffic. This hybrid algorithm is also employed to tune the features selected which is assigned as initial weight vectors for both BPN and MLP intrusion detection system (IDS) models. These selected optimally tuned features are trained using 10fold cross validation technique and the number of hidden neurons is fixed using thumb rule. After training, the hybrid BPN-MLP neural network IDS model is validated on test dataset and its performance is validated using performance metrics such as accuracy, precision, sensitivity, specificity and F1 score. Findings: The proposed hybrid HHO-PSO BPN and HHOPSO MLP IDS model has achieved detection accuracy of 97.08% and 97.74% with F1 score of 0.9743 and 0.9800 respectively. Novelty: In ANN based intrusion detection schemes, the stochastic nature of model parameters is an important problem of concern. To handle this issue, a hybrid swarm intelligent algorithm called Harris hawks optimization particle swarm optimization (HHOPSO) is proposed to tune the model parameters, so that the network performance is enhanced.