The amount of malware has been rising drastically as the Android operating system enabled smartphones and tablets are gaining popularity around the world in last couple of years. One of the popular methods of static detection techniques is permission/feature-based detection of malware through the AndroidManifest.xml file using machine learning classifiers. Ignoring important features or keeping irrelevant features may specifically cause mystification to classification algorithms. Therefore, to reduce classification time and improve accuracy, different feature reduction tools have been used in past literature. Community detection is one of the major tools in social network analysis but its implementation in the context of malware detection is quite rare. In this article, the authors introduce a community-based feature reduction technique for Android malware detection. The proposed method is evaluated on two datasets consisting of 3004 benign components and 1363 malware components. The proposed community-based feature reduction technique produces a classification accuracy of 98.20% and ROC value up to 0.989.

中文翻译：

---

基于社区的Android恶意软件特征选择方法

随着最近几年启用Android操作系统的智能手机和平板电脑在全球范围内的普及，恶意软件的数量急剧增加。静态检测技术的一种流行方法是使用机器学习分类器通过AndroidManifest.xml文件对恶意软件进行基于权限/特征的检测。忽略重要特征或保留不相关特征可能会特别引起分类算法的神秘化。因此，为了减少分类时间并提高准确性，在过去的文献中使用了不同的特征减少工具。社区检测是社交网络分析中的主要工具之一，但是在恶意软件检测的背景下实施社区检测的情况却很少。在这篇文章中，作者介绍了一种基于社区的功能约简技术，用于Android恶意软件检测。该方法在包含3004个良性组件和1363个恶意软件组件的两个数据集上进行了评估。所提出的基于社区的特征约简技术可产生98.20％的分类精度，ROC值高达0.989。