Abstract:

The rapid growth of the Internet of Things (IoT) in smart cities produces vast and diverse network traffic, demanding reliable device classification methods for security and management. This paper reviews existing approaches to IoT device classification based on network traffic, focusing on feature extraction and algorithmic techniques. Feature extraction methods include packet-based, flow-based, and deep-learning approaches. Packet analysis offers precision but struggles with encrypted traffic, while flow-level and deep learning methods improve scalability and adaptability. The review highlights supervised, unsupervised, and deep learning algorithms, with accuracy reaching up to 99%. Key research directions include experimental validation, edge computing integration, and hybrid traffic–hardware classification models to enhance robustness in smart city environments.