A three-level deep learning intrusion detection system for IoT network
Abdoulaye Diallo, Lionel Affognon, Cherif Diallo, Eugène C Ezin
Informations
- Revue / Journal
- 2023 International Conference on Electrical, Communication and Computer Engineering (ICECCE)
- ISBN
- 979-8-3503-6969-4
Résumé
Security stands out as one of the most significant challenges confronting the Internet of Things. Within the existing literature, various solutions have been proposed, with intrusion detection systems playing a prominent role and offering numerous approaches to address security concerns. This paper introduces a three-level intrusion detection system based on deep learning. At the first level, an autoencoder is deployed to detect intrusion presence. Moving to the second level, another autoencoder is utilized to classify the intrusion as either a known or new attack when an intrusion occurs. In the case of a recognized attack, the third level comes into play. At this stage, a Multi-Layer Perceptron (MLP) serves as a multi-class classifier, enabling precise identification of the ongoing attack type. The entire system is trained and evaluated using the MQTTSet dataset. Comprehensive tests are conducted at each of the three levels. The results obtained demonstrate an accuracy of 99%, 99.60%, and almost 100% respectively at level 1, level 2, and level 3.