Hyoseong Park, Dongkyoo Shin, Dongil Shin, Journal of Internet Computing and Services, Vol. 26, No. 1, pp. 149-155, Feb. 2025
10.7472/jksii.2025.26.1.149, Full Text:  HTML
Keywords: LSTM, Autoencoder, Network intrusion detection, CIC-IDS2018

Abstract

Attack detection methods based on packet signatures are most used to detect intrusions into networks, but they are vulnerable to unknown attacks. Therefore, there are many models that use machine learning to learn network packets and determine the presence of unknown attacks. In this study, we designed a Long Short-Term Memory Autoencoder (LSTM autoencoder) model, which shows good performance in learning time series data, to learn normal data, test the model on mixed data containing normal and intrusive data, and calculate the reconstruction loss. Based on this reconstruction loss value, a network intrusion detection model is studied to perform intrusion detection by obtaining a threshold value. The final experimented combined model shows a high performance of 98% accuracy and F-1 score of 98%.

Statistics
Cite this article