Intrusion detection using a combination of one-dimensional convolution and GRU

Abstract

Intrusion detection plays an important role in ensuring network information security. Traditional machine learning technology does not work well enough with massive data and various intrusion classes, and detection accuracy turns unsatisfied with unknown intrusions. This paper proposes a new network intrusion detection model (Conv1d-GRU) that combines one-dimensional convolution and GRU for multi-class intrusion detection scenarios, and improves data imbalance by weighting the samples of each category. NSL-KDD as an improved version of the KDD CUP99 dataset is selected for our intrusion detection system. Experimental results on this dataset show that the proposed deep learning method is superior to present intrusion detection methods based on machine learning and deep learning, and has better feature representation learning and classification capabilities.