Network Traffic Identification Based on Improved EM Algorithm

Abstract

With the continuous increase and complexity of network traffic, traditional network traffic recognition technology is facing numerous difficulties, especially in dealing with outlier data and improving recognition accuracy. Therefore, an improved expectation maximization algorithm based on the constraint matrix Z and Tsallis entropy is proposed. The core goal of this algorithm is to accelerate the convergence of classification and improve accuracy. Furthermore, to enhance the classification accuracy, the spatial expectation maximization algorithm is introduced, which innovatively converts the sample mean and covariance matrix into $L_{1}$ -median and modified rank covariance matrix. According to the experimental data, the recall rate of the original expectation maximization algorithm is only 74%. However, the recall rate of the spatial expectation maximization algorithm in the Attack service has significantly increased to 85%. In other tests, such as Www and Peer-to-peer services, the recall rate has also significantly improved, increasing from 96% and 95.3% to 97.7% and 96.1%, respectively. These experimental results highlight the superior robustness of the spatial expectation maximization algorithm in handling outlier data. It further proves the outstanding performance in improving the accuracy of network traffic recognition. This research has brought significant innovation and potential practical value to the network traffic identification.

Keywords

• Em algorithm
• network traffic identification
• Gaussian mixture model
• Tsallis entropy
• restriction matrix