A Efficient Network Traffic Classification Method based on Combined Feature Dimensionality Reduction

Network traffic classification is the key process of malicious traffic identification. It determines the type of network traffic from a large amount of network traffic data according to the corresponding features of various traffic so that it can identify the vulnerability attacks. Amone them, the machine learning-based network traffic classification method is widely used because of its high precision and easy deployment. However, when the dimensionality of network traffic data is high, the effectiveness of this method is greatly affected, and it is very time-consuming. Feature dimensionality reduction is a common method to solve this problem, but the data processed by the existing feature dimensionally reduction methods still have many irrelevant features and high-dimensional problems are not addressed.This paper proposes an efficient network traffic classification method based on combined feature dimensionality reduction. It first combines the ReliefF algorithm and the Principal Component Analysis (PCA) feature extraction algorithm to determine the key features and then apply it to the BanlaceBaggingRF classification model to classify and identify network traffic. A large amount of experimental results show that the combined feature dimensionality reduction method can effectively remove the redundant features in network traffic data and address high-dimensional space problems, and the accuracy of classification and recognition of the BanlanceBaggingRF model is also improved about 2% to 10% compared with the latest-of-the-art.