Date

2022-9-14

Author

Tüzün, İsmail

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

271
views

63
downloads

Cite This

As the number of devices using the Internet increases, the network attacks that these devices are exposed to also diversify. Identifying network attack types from network packets is important to prevent the damage of the attack and to minimize it in cases where it cannot be prevented. Classical machine learning methods and deep learning methods need a lot of data to get successful results. Unfortunately, preparing and labeling large amounts of data is costly in current conditions. This cost is mostly due to the training of the experts who will do the labeling process, the difficulty of generating attack environments, and the complexity of attacks. This study examines the problem of classifying network attacks with limited data in the learning process by applying few-shot learning methods. To investigate the problem, we generate three different datasets using previously labeled large datasets including CIC-IDS2017 and UNSW-NB15. We apply three promising approaches, where two of them are based on Prototypical Networks, and one of them is based on Relation Networks.

Subject Keywords

Cybersecurity, Network attack classification, Few-shot learning, Network intrusion detection, Meta-Learning framework

URI

https://hdl.handle.net/11511/99604

Collections

Graduate School of Natural and Applied Sciences, Thesis