A Comparative Assessment of Deep Learning Algorithms in Network Intrusion Detection System

Network Intrusion Detection System (NIDS) is crucially important to protect computer network systems from unauthorized access, malicious attacks, and different kinds of cyber threats. While the cyber threat evolves, the original threat from traditional detection methods can't make rapid and accurate detection. Deep Learning (DL) methods have recently attracted more and more attention as a promising approach to improve the capability of NIDS in processing large amount of network data and learning complicated patterns. This article presents a comparative analysis of deep learning algorithms adopted in NIDS. In this scope, the authors want to find the best models for NIDS, according to their classification accuracy, false positive/negative rates and computational cost. We test Convolutional, Recurrent and Deep Neural Network (DNN) models. This work also particularly studies the potentials of feature engineering, dataset preprocessing and hyperparameter tuning that can be utilized to improve model performance. The results provide promising clues on how these deep learning architectures might lead to more robust and adaptive IDS for various network security tasks.