Design and Development of Deep Learning Architecture for Network Intrusion Detection System

Abstract

newline Abstract newlineData Security is a fundamental pillar in modern ICTs, aiming to avoid data newlinebreaches. In this context, malware detection, network intrusion and software newlinevulnerability analysis are related problems. To fight against the increasing newlinenumber of attacks, their variability and sophistication, machine learning-based newlinesolutions, which can model the previous problems using sequence-based newlineneural networks and Natural Language Processing (NLP) concepts have newlineemerged as solution and it is being increasingly adopted by companies and newlineinstitutions. In particular, deep learning have started to show their impressive newlineperformance on the security domain. newlineThis research focuses on the comprehensive methodology for developing a newlinerobust Network Intrusion Detection System (NIDS) using the CICIDS2017, newlineHIKARI-2021, and 5G-NIDD datasets to assess the efficacy of the suggested newlinemodels across different network environments and attack scenarios, newlineemploying traditional ML models and advanced DL architectures. Deep newlinelearning models, particularly those incorporating dense and dropout layers, newlineexhibit superior performance in handling large and complex datasets. The newlineresearch emphasizes the critical role of feature engineering, data pre newlineprocessing, and model optimization in accurately identifying and classifying newlinenetwork intrusions. The performance for each dataset is evaluated based on the newlineconfusion matrix statistical parameters, accuracy, recall, precision, and f-score newlinewith evaluation time required for the training and testing process. newlineThe deep learning model based on SD-RNN demonstrated superior newlineperformance in detecting sequential and time-dependent network anomalies. newlineThis was particularly evident in the analysis of the CICIDS2017 and 5G-NIDD newlinedatasets, where network traffic behaviours over time played a critical role in newlinedistinguishing between normal and malicious activities. Overall, this research newlinecontributes to the field by demonstrating the effectiveness of The SD-RNN newlineapproach was able to capture the temp