Abstract

With the rapid growth of computer networks, cyber threats have become more sophisticated and effective, making it challenging to detect them accurately. One solution is to use an intrusion detection system (IDS). This paper reviews the use of deep learning methodologies for active cyber threat detection, focusing on classification methods such as autoencoders, RNNs, BiLSTMs, DNNs, CNNs, and XGBoost, as well as deep learning models like sequence-to-sequence and neural networks with convolutional models. The study also examines the datasets used by researchers and the evaluation criteria for IDS techniques, as well as the class imbalance problem and its impact on relevant research. This review aims to provide guidance for scholars and industry professionals interested in deep learning-based IDSs.