深度学习方法存在数据集庞大、特征维度高等问题，这些问题导致其时间效率较

Deep learning methods have problems with huge datasets and high feature dimensions, which lead to low time efficiency, high computational costs, and high model training costs. From a practical point of view, Tor traffic identification should minimize feature selection time and model training time while maintaining high accuracy. Therefore, this paper proposes an identification method that can effectively deal with the sparse Tor traffic problem and has strong stability, which can obtain a better identification effect while increasing the limited computational overhead.

The deep learning method has the problems of large data set and high feature dimension, which lead to its low time efficiency, high computing cost and high cost of model training. From a practical point of view, tor traffic recognition should have high accuracy and minimize feature selection time and model training time. Therefore, this paper proposes an effective and stable identification method to deal with the sparse problem of tor traffic, which can increase the limited computational overhead and get a better identification effect.

Deep learning method has some problems, such as huge data set and high feature dimension, which lead to low time efficiency, high calculation cost and high model training cost. From a practical point of view, Tor traffic identification should have a high accuracy while minimizing the time of feature selection and model training. Therefore, this paper proposes an effective and stable identification method to deal with the sparse Tor traffic problem, which can increase the limited computational cost and get a better identification effect.