Abstract

The rise of sophisticated cyber threats, particularly Denial of Service (DoS) and User to Root (U2R) attacks, underscores the need for advanced intrusion detection systems (IDS) capable of real-time threat detection and mitigation. Traditional IDS methods often struggle to recognize complex, temporal patterns in network traffic, presenting an opportunity for deep learning models to improve accuracy and adaptability. This study investigates the performance of various recurrent neural network (RNN) architectures—namely Long Short-Term Memory (LSTM), Gated Recurrent Units (GRU), and Bidirectional LSTMs—for the detection of DoS and U2R attacks. By evaluating these architectures across key metrics such as accuracy, precision, recall, and computational efficiency, we aim to identify the most effective model for real-time deployment in network security systems. Using the NSL-KDD dataset, we conduct experiments to compare model performance, fine-tune hyperparameters, and assess each architecture’s suitability for real-time application.