一种基于图像特征的移动流量分类方法

Classification method of mobile traffic based on image features

随着移动设备的快速发展和应用激增,其产生的移动流量也迅猛增加且众多操作系统皆存在着巨大的安全风险,能够从巨大的网络流量中有效地区分出来自移动端的流量并识别其操作系统对后续的移动安全的分析有着重大的安全意义。基于传统特征的流量分析技术存在着因过分依赖特征选择而导致无法稳定有效地分类移动流量的问题,提出了一种基于图像特征的移动流量分类方法。该方法将流量样本进行可视化转换成灰度图像,从而提取其图像的GLCM特征进行分类。实验结果表明,该方法较传统方法精确率最高提升22.4%,有效地解决了传统方法的特征选择以及没有良好的扩展性等问题。此外,研究了流量研究粒度(flow到stream)、分类粒度(二分类到多分类)和数据集的均衡性(均衡与不均衡)对移动流量检测方法的影响,结果表明分类粒度对分类准确率的影响极小,准确率最大降低2.6%。该实验结果进一步说明了提出方法的扩展性,能够有效地用于后续移动流量的安全研究。

The mobile traffic is increasing rapidly besides exists great risk in many operating systems because of the rapid development and application explosion of mobile devices.Therefore,it is of great significance to effectively distinguish mobile traffic and identify the operating system from huge network traffic for mobile malicious traffic analysis.Traffic analysis technology based on traditional features has the problem of relying too much on feature selection to classify mobile traffic steadily and effectively.Due to above reasons,this paper proposed a mobile traffic classification method based on image features,which visualized the traffic samples,transformed them into gray images and extracted the GLCM features of the images for classification.The experimental results show that the accuracy of the proposed method is 22.4%higher than that of the traditional method,which effectively solves the problems of feature selection and lack of good scalability of the traditional method.In addition,this paper studied the influence of traffic granularity(flow vs stream),classification granularity(two classifications vs multiple classifications)and the equilibrium of datasets(balance vs imbalance)on mobile traffic classification methods.The research shows that the influence degree of classification granularity has little effect on the proposed method,and the highest reduced accuracy is only 2.6%.The experimental results further illustrate the expansibility of this method,which can be effectively used in the security research field of subsequent mobile traffic.