KDTC: A Lightweight Traffic Classification Method Based on Knowledge Distillation

In this study, a lightweight traffic classification method based on knowledge distillation (KDTC) is proposed. The method is based on the knowledge distillation technique and aims to solve the current challenges of network traffic classification. With the proliferation of mobile devices and applications, network traffic has increased dramatically and traditional classification methods are no longer applicable. KDTC achieves efficient and accurate traffic classification by transferring knowledge from the complex teacher model to the more compact student model. Classification accuracy, especially for complex samples, is improved by designing the distillation loss function. The lightweight MobileNetV3 model is adopted as the student model, which optimizes the model efficiency and meets the demand for fast response and low resource consumption in real-time environments. The introduction of modulation coefficients reduces the risk of overfitting during student model training and improves the generalization ability of the model. Experimental results show that the KDTC significantly optimizes computational efficiency and reduces resource consumption while maintaining model accuracy. This provides a new technical path for deploying efficient and accurate deep learning models in resource-constrained environments, which is important for traffic classification in mobile devices and edge computing.