IoT anomaly detection and attack identification using smart traffic classification techniques
Cherif Diallo
Informations
- Revue / Journal
- 2022 7th International Conference on Frontiers of Signal Processing (ICFSP)
- Éditeur
- IEEE
- ISBN
- 978-1-6654-8158-8
Résumé
Attacks in IoT networks are increasingly numerous and frequent. Therefore, knowing how to extract abnormal traffic in order to identify potential attacks seems to be a task of primary importance. Here, we first seek to detect anomalies in smart home IoT network traffic through binary classification. Then we try to identify the attacks contained in the anomalous traffic by using learning models based on multiclass classification. In each of these two cases, we used two different datasets to study and compare four learning models based on Decision-Tree (Adaboost), Random Forest, KNN (K-Nearest Neighbors) and a Neural Network model. The results show that our models based on Decision-Tree and Random Forest are more efficient both in the detection of anomalies and in the identification of attacks. This is also the case both in predictions and in behavior with new data.