基于深度森林与CWGAN-GP的移动应用网络行为分类与评估

针对目前移动应用数目庞大、功能复杂,并且其中混杂着各式各样的恶意应用等问题,面向Android平台分析了应用程序的网络行为,对不同类别的应用程序设计了合理的网络行为触发事件以模拟网络交互行为,提出了网络事件行为序列,并利用改进的深度森林模型对应用进行分类识别,最优分类准确率可达99.03%,并且其具有高精确率、高召回率、高F1-Score和低训练时间的特点。此外,为了解决应用样本数量有限且数据获取时间开销大等难题,还提出了一种使用CWGAN-GP的数据增强方法。与原始生成对抗网络相比,该模型训练更加稳定,仅需一次训练即可生成指定类别的数据。实验结果表明,在加入生成数据共同训练深度森林模型后,其分类准确率提高了9%左右。

Classification and Evaluation of Mobile Application Network Behavior Based on Deep Forest and CWGAN-GP

In view of the problems that the large number and complex functions of mobile applications,and mixed with a variety of malicious applications,this paper analyzed the network behavior of applications for Android platform,and designed reasonable network behavior trigger events for different types of applications to simulate network interaction behavior.Based on the network event behavior sequence,the improved deep forest model is used to classify and identify applications.The optimal classification accuracy can reach 99.03%,and it has high accuracy,high recall rate,high F1-Score and low training time.In addition,in order to solve the problems of limited number of application samples and high time cost of data acquisition,a data enhancement method using CWGAN-GP was proposed.Compared with the original generative adversarial network,the training of the model is more stable,and the data of specified categories can be generated by only one training.The experimental results show that the classification accuracy is improved by about 9%after joining the generated data to train the deep forest model together.