Performing Multiclass Classification on UNSW-NB15 Dataset by Applying Machine Learning Approach on Intrusion Detection System

It is still difficult to accurately identify computer network assaults in this era of technology. This is because hackers have recently tried to conceal the intrusion detection system (IDS) by changing packet contents. In addition to exploring novel attack kinds, defenses, and recent scientific investigations in this field, this paper examines intrusion detection technologies, methodology, and tactics. Additionally, freely available datasets this paper uses, UNSW NB15 dataset, popular IDS technologies, and the advantages and disadvantages of certain IDS are extensively covered. Deep learning (DL), a subset of machine learning (ML), excels in intrusion detection and is based on very accurate, complicated neural networks (NN). A CNN and BiLSTM combination is proposed in this study work, enabling the efficacy of their mixture in routine identification following automated data categorization. Because LSTM is an extra recurrent neural network (RNN) integration, temporal sequences may be easily added to it. On this dataset (UNSW NB15), pooling layer (PL) reshape is used to lower the model’s parameters across the connected layer. The dataset contains 49 features information acquired from various attack-based activities performed within the proposed model. The accuracy of the hybrid models described in this study, Convolutional Neural Network Bi-Directional Long Short Term Memory (GRU-BiLSTM), has now been evaluated using a variety of network assaults. In comparison with the previous method, the new one obtains a greater degree of accuracy of 81.94%.