Unbalanced Data Oversampling Method for Traffic Multi-classification in Convergence Network

Abstract

At present, classification models are widely used in network traffic anomaly detection. Researchers have found that the imbalance of network traffic data sets affects the result of anomaly detection in convergence network. For imbalanced data, researchers have proposed many solutions for dichotomous datasets, and there are often multi-class datasets in network traffic datasets, that is, there are often many different types of anomalous data in network traffic data, and the frequency of occurrence between different anomalous data varies greatly. To address the problems of complex data information and different important features of each class that often exist in such multi-class unbalanced network traffic data, this paper proposes a new oversampling technique, which firstly performs sample selection based on the information entropy of minority class samples, followed by the decomposition of eigenvalues of the data using principal component analysis, making the features uncorrelated with each other, and then synthesizes the data. Finally, we validate the effectiveness of our method on handling multi-class network traffic datasets.