Abstract

Due to a rapid development in the field of information and communication, the information technologies yielded novel changes in both individual and organizational operations. Therefore, the accessibility of information became easier and more convenient than before, and malicious approaches such as hacking or spying aimed at various information kept increasing. With the aim of preventing malicious approaches, both classification and detecting malicious traffic are vital. Therefore, our research utilized various deep learning and machine learning models for better classification. The given dataset consists of normal and malicious data and these data types are png files. In order to achieve precise classification, our experiment consists of three steps. Firstly, only vanilla CNN was used for the classification and the highest score was 86.2%. Second of all, for the hybrid approach, the machine learning classifiers were used instead of fully connected layers from the vanilla CNN and it yielded about 87% with the extra tree classifier. At last, the Xception model was combined with the bidirectional GRU and it attained a 95.6% accuracy score, which was the highest among all.

How to Cite
JOO, HAYOUNG CHOI, CHANGHUI YUN, MINJONG CHEON, HanSeon. Efficient Network Traffic Classification and Visualizing Abnormal Part Via Hybrid Deep Learning Approach : Xception + Bidirectional GRU. Global Journal of Computer Science and Technology, [S.l.], jan. 2022. ISSN 0975-4172. Available at: <https://computerresearch.org/index.php/computer/article/view/2070>. Date accessed: 15 aug. 2022. doi: https://doi.org/10.34257/GJCSTHVOL21IS3PG1.