Anomaly detection in Internet of Things equipment traffic with the approach of combining deep learning models

Document Type : Original Article

Authors

1 PhD student, Allameh Tabatabai University, Tehran, Iran

2 Researcher, Qom Higher Education Institute, Qom, Iran

3 Ph.D. student, Faculty of Computer Science, Islamic Azad University, North Tehran Branch, Tehran, Iran

Abstract

Network traffic classifiers play a critical role in network monitoring systems by detecting anomalies in network flows based on communication features. This is particularly significant for managing and monitoring Internet of Things (IoT) networks. In this study, a novel method for network traffic classification is proposed, leveraging a combination of deep learning models tailored for IoT traffic. Experimental results demonstrate that the hybrid CNN+RNN-2 model achieves an accuracy of 83.58%, outperforming standalone and other hybrid models. By combining local feature extraction through Convolutional Neural Networks (CNN) and temporal dependency analysis via Recurrent Neural Networks (RNN), the proposed model effectively identifies complex patterns and improves detection accuracy. Furthermore, the results indicate that the CNN+RNN-2 model surpasses traditional supervised, semi-supervised, and unsupervised methods without requiring manual feature engineering. The use of deep learning, with its capability for automatic feature extraction and learning complex patterns, provides a significant advantage over conventional artificial intelligence techniques.

Keywords

Main Subjects

Smiley face

 

Files

History

  • Receive Date: 23 April 2025
  • Revise Date: 25 May 2025
  • Accept Date: 08 June 2025
  • Publish Date: 22 June 2025

Share

How to cite

Statistics