Deep Learning-based Techniques for Intrusion Detection Systems
Nafay Rizwani, Akhtar Jamıl, Alaa Ali Hameed · 2024
In the evolving landscape of cybersecurity, effective Intrusion Detection Systems (IDS) are essential for protecting against increasingly sophisticated cyber threats. This study explores the efficacy of various neural network models in detecting network intrusions, utilizing the widely recognized UNSW NB15 dataset. Our research encompasses a comparative analysis of Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM) networks, Gated Recurrent Units (GRU), Artificial Neural Networks (ANN), and DenseNet. Each model was meticulously tailored and evaluated for binary (attack or not) and multiclass (type of attack) classification tasks. In addition, to address the challenge of UNSW-NB 15 dataset, we implemented strategic data preprocessing techniques, including scaling and over/under sampling techniques, aiming to enhance model accuracy and generalizability. The models’ performances were rigorously assessed using key metrics for binary and multiclass classification. These metrics provided a comprehensive understanding of each model’s strengths and weaknesses in the context of IDS. Furthermore, our research highlights the impact of various data preprocessing strategies, establishing benchmarks for future advancements in deep learning-based IDS.