Abstract: With more and more widespread use of smartphones, malwares have become increasingly complex and large-scalely. As a free and open source system, Android has currently surpassed other mobile platforms to become the most popular operating system, so that the number of the Android platform malware has also been significantly increased. Focusing on the security issues of the software for the Android platform, this paper proposes an Android malware detection method based on multifeature collaborative decision. This method mainly bases on the analysis of the Android application, and then the feature attributes are extracted, the models according to machine learning are built. Lastly the classification algorithms are used to determine whether the application is malware. Experimental results show that using the proposed method to classify Android application data set has better assessments indicators than the indicators using other classifiers. Therefore, the method based on multi-feature collaborative decision approach to detect malicious software on Android applications can be made effective for detecting unknown malicious nature of the applications, and can avoid damage caused by malicious applications for the users.

Key words: Android platform, malware, multifeature collaborative decision, machine learning

摘要： 由于智能手机使用率持续上升促使移动恶意软件在规模和复杂性方面发展更加迅速。作为免费和开源的系统，目前Android已经超越其他移动平台成为最流行的操作系统，使得针对Android平台的恶意软件数量也显著增加。针对Android平台应用软件安全问题，提出了一种基于多特征协作决策的Android恶意软件检测方法，该方法主要通过对Android 应用程序进行分析、提取特征属性以及根据机器学习模型和分类算法判断其是否为恶意软件。通过实验表明，使用该方法对Android应用软件数据集进行分类后，相比其他分类器或算法分类的结果，其各项评估指标均大幅提高。因此，提出的基于多特征协作决策的方式来对Android恶意软件进行检测的方法可以有效地用于对未知应用的恶意性进行检测，避免恶意应用对用户所造成的损害等。

关键词: Android平台, 恶意软件, 多特征协作决策, 机器学习