Fusion of Federated Learning and Improved ResNeXt for Encrypted Traffic Identification

To solve the problem of low efficiency in extracting features and key protocol fields in traditional machine learning-based classification techniques and the vulnerability to sensitive data leakage, federated learning and improved ResNeXt-based encrypted traffic classification model, FL-ResNeXt, is proposed to achieve the identification of encrypted traffic. First, the dataset is preprocessed at the packet level and converted into grayscale images. Then, during the training process, data is distributed, extraction, and aggregated to enable joint training data provision from multiple parties while protecting user privacy. In this process, the improved ResNeXt network effectively improves the model's ability to capture complex encrypted traffic features through its optimized block structure and parallel stacking of the same topology design while retaining more fine-grained information and enhancing classification accuracy. Experimental results show that this method achieves better classification accuracy than advanced encrypted traffic classification models and ensures data security.