Study and development of network based intrusion detection system

Abstract

The rapid growth of internet usage and the complexity of modern networks have newlinesignificantly increased the risk of cyberattacks, necessitating efficient and newlineintelligent intrusion detection systems (IDS). This thesis presents the study and newlinedevelopment of a network-based IDS leveraging advanced deep learning newlinealgorithms, with a focus on improving anomaly detection accuracy, reducing newlinefalse alarms, and addressing challenges of class imbalance. The research begins newlinewith an exploration of state-of-the-art anomaly detection methods, including newlineensemble learning, federated learning, and statistical flow feature analysis. A newline newlinehybrid approach combining recurrent neural networks (RNN) with correlation- newlinebased feature optimization is proposed to enhance temporal dependency learning newline newlinein network traffic. Further, a novel CNN-GRU hybrid model is introduced, newlineintegrating the feature extraction capability of convolutional networks with the newlinesequential learning strength of gated recurrent units (GRU). Extensive newlineexperiments on the CICIDS 2017 dataset validate the effectiveness of these newlinemodels. Additionally, the role of GRU in IDS is evaluated through comparative newlineanalysis with traditional RNN architectures, highlighting its efficiency in newlinehandling long-term dependencies. To mitigate class imbalance, the Synthetic newlineMinority Oversampling Technique (SMOTE) is employed, significantly newlineimproving classification performance across minority attack categories. The newlineproposed methodologies demonstrate superior detection rates, robustness against newlineevolving threats, and scalability for real-world deployment. The contributions of newlinethis work lie in developing optimized deep learning architectures, addressing newlinedataset imbalance, and advancing the state-of-the-art in anomaly-based intrusion newlinedetection systems. The thesis concludes with insights into practical implications newlineand future directions, including real-time IDS integration and the use of advanced newlinedeep learning paradigms. newline