Massive IoT network traffic analysis using ML and DL methods: an empirical evaluation

Massive IoT network traffic analysis using ML and DL methods: an empirical evaluation

연구 분야: Networking

학회: The Journal of Supercomputing

As the proliferation of Internet of Things (IoT) devices continues, securing IoT networks against cyber threats becomes increasingly vital. Detecting and mitigating attacks targeting interconnected IoT devices pose significant challenges. Machine learning (ML) and deep learning (DL) techniques offer promising solutions for detecting various cyber threats in IoT networks. However, empirical evaluations comparing the performance of different ML and DL methods for IoT network attack detection remain limited. This study provides a thorough empirical assessment of ML and DL methods using four standard datasets, NBaIoT, Bot-IoT, Edge-IIoT, and CICIoT2023, for IoT network attack detection. Additionally, we extend our experimental analysis to check the significance of feature selection using mutual information-based feature selection. Various ML and DL models undergo training and evaluation using this dataset, considering different feature representations and network architectures. Performance is assessed using metrics such as accuracy, precision, recall, F1-score, training time, prediction time, Area under the ROC Curve (AUC) score, Matthews correlation coefficient (MCC), and Kappa score. Our experimental findings underscore the efficacy of ML and DL methods in identifying attacks on IoT networks. We discuss the balance between computational overhead and detection accuracy. This empirical evaluation contributes valuable insights into the performance of ML and DL methods for IoT network attack detection, informing future research directions and practical implementations in securing IoT ecosystems.