ConFedDI: Efficient IoT Device Identification via Horizontal Federated Learning and Convolutional Neural Networks

ABSTRACT With the exponential growth of Internet of Things (IoT) devices, device identification has become a critical need for ensuring network security and Quality of Service (QoS). However, traditional centralized identification methods generally face issues such as high risk of privacy leakage, substantial communication costs, and poor model generalization. To address these challenges, this paper proposes an IoT device identification method based on Horizontal Federated Learning (HFL) and Convolutional Neural Networks (CNNs), named ConFedDI. Leveraging the training mechanism of HFL, this method allows multiple edge devices to train models locally, requiring only the upload of model parameters rather than raw data, thereby effectively protecting data privacy and significantly reducing communication costs. At the edge device level, ConFedDI extracts features from device behavioral traffic packet headers and utilizes an improved lightweight AlexNet network to automatically learn complex feature representations at different levels, enabling efficient device classification. To tackle the prevalent issue of statistical heterogeneity in distributed learning scenarios, the method incorporates a weighted cross‐entropy loss function and a stochastic gradient descent (SGD) optimizer with momentum. Comprehensive experiments on two public datasets demonstrate that: ablation studies verify that ConFedDI outperforms the M3CNN model in performance while incurring lower communication costs; comparative experiments show that ConFedDI significantly surpasses the HFedDI and IoTDevID methods in identification performance, with F1‐scores reaching 99.44% and 99.75% on the two datasets, respectively.