基于马氏距离的重采样方法在流量识别中的应用

Application of Resampling Method Based on Mahalanobis Distance in Traffic Identification

针对网络流量识别中的多分类数据分布不均衡的问题,本文提出了一种基于马氏距离的重采样方法。首先,将网络流量数据进行零均值化处理并转换至主成分空间;再根据少数类样本数据到集合中心点之间的马氏距离对其进行新样本的生成;之后将新生成的样本数据转换至原始空间并进行逆零均值化处理;最后返回所有新生成的样本数据。使用剑桥大学公共网络流量数据进行流量分类实验,实验结果表明该方法能够有效提升少数类的识别准确率,并且比现有的重采样方法和成本敏感方法能够获得更好的分类效果。

Aiming at the problems of multi-class imbalance of data distribution in traffic identification,this paper proposed a novel resampling method based on Mahalanobis distance.First,the network traffic data is normalized and transformed to the principal component space;second,a new sample is generated for a minority class based on the Mahalanobis distance from the samples to the center point of the data set;third the newly generated sample is then transfomed to the original space and performed an anti-normalization process;and finally,all the new samples are returned to original data set.The public Internet traffic traces of Cambridge University is used for traffic classification experiment,the results show that the proposed method can effectively improve the accuracy of the minority classes in traffic data sets,and it can obtain better classification performance than the existing resampling methods and cost-sensitive methods.