摘要
Android操作系统作为当前最为流行的智能手机应用平台,但受到各种恶意软件的攻击。目前已有研究基于抽取的恶意软件特征来构建机器学习分类器完成恶意软件检测,但实际应用中我们通常仅能获取少量的标记数据和大量的无标记数据,因此如何有效地利用少量的标记数据集和大量的无标记数据成为当前研究一个挑战。为此,该文提出了一种基于半监督学习的恶意软件检测方法。首先,我们选取了一些特征表征隐藏在Android恶意软件中的恶意代码;然后,我们基于少量的标记数据和大量的无标记数据构建半监督分类模型,通过类EM迭代算法优化朴素贝叶斯分类器;最后,通过公开数据集Virus Share验证算法的有效性。
Android, one of most popular open source mobile operating systems, is confronted with lots of malicious threats. Al- though many studies have been presented to malware detection, which do not agree with the fact that, in practice, we are often given only a few labeled but a majority of unlabeled data. How to effectively utilize a small number of labeled data sets and a large number of unlabeled data has become a challenge problem for android malware detection. In this paper, we propose a semi-su- pervised learning-based method to detect the android malware (SSAMD for short). Firstly, we select some features combination of permissions and resources. Secondly, a semi-supervised learning system is employed for the categorization of both labeled and unlabeled data. Extensive experiments on VirusShare datasets demonstrate the effective of our presented algorithm.
作者
陈志刚
王青国
CHEN Zhi-gang1, WANG Qing-guo2 (1.State Grid Jiangsu Electric Power Company, Nanjing 210024, China; 2. Jiangsu Electric Power Information Technology Co Ltd., Nanjing 210013, China)
出处
《电脑知识与技术》
2017年第12期265-268,共4页
Computer Knowledge and Technology
基金
江苏省科技支撑计划工业项目(BE2014141)