A Hybrid Penguin Search and Deep Neural Network Approach for Intrusion Detection with Parameter Optimization via Levy Gray Wolf Algorithm

Abstract

To protect a network from advanced cyberattacks, it needs to use smart and strong intrusion detection systems (IDS). Current IDS have a hard time working because network traffic data has a lot of dimensions and is uneven, and traffic is getting worse. Traditional IDS often stop working too because they have features that are not needed, they cannot be made bigger, and they do not adjust their settings well. So, it is very important to find a small set of features that are very different from each other and have been optimized for learning. This study suggests a new hybrid model that would use a feature selection algorithm based on the Penguin Search Algorithm (PSA DNN) and a Deep Neural Network (DNN). The goal is to improve the detection and classification of intrusions. To begin making the dataset more consistent, we use z-score normalization. We look at five benchmark datasets, NSL-KDD, ISCX 2012, UNSW-NB15, KDD Cup 1999, and CICIDS 2018, to find out which features are the most important. Then, PSA is used to find these features and a DNN is trained on these features so that it can sort traffic correctly. We change the hyperparameters of the DNN with the Levy Gray Wolf Optimization (LGWO) algorithm to make it as good as possible at classifying things. The suggested PSA-DNN model is more accurate at both segmentation and classification on all the datasets it was tested on. When it comes to precision, recall, F1 score, and overall accuracy, it works better than traditional methods. On the CICIDS test, the model gets an average accuracy of 99.23%, a precision of 98.9%, a recall of 99.1%, and an F1-score of 99.0%. Because it has low false positive rates and high detection rates, the model works well in the real world where threats change quickly.