DroidDeepLearner: Identifying Android malware using deep learning
DroidDeepLearner: Identifying Android malware using deep learning
复制标题
DOI：
10.1109/sarnof.2016.7846747
复制
发表时间：
2016-09
期刊：
影响因子：
--
通讯作者：
Zi Wang;Juecong Cai;Sihua Cheng;Wenjia Li
中科院分区：
文献类型：
--
作者：
Zi Wang;Juecong Cai;Sihua Cheng;Wenjia Li
随着Android应用程序的激增，手机用户遇到的恶意应用程序（恶意软件）也在上升，因为恶意软件的作者利用Android平台系统的漏洞来访问个人或敏感信息，通常是为了获得经济利益。为了维护安全完整性和维护用户信心，恶意软件检测领域已经研究了各种方法。随着恶意软件越来越有能力通过使用代码混淆来隐藏其恶意意图，恶意软件检测技术必须跟上恶意软件变化的步伐。目前，Android平台现有的恶意软件检测方法大多采用语义模式匹配的方法，这种方法效率很高，但受限于计算机之前遇到的情况。然而，当涉及到识别以前从未处理过的恶意应用程序时，它们的性能会显著下降。在本文中，我们提出了DroidDeepLearner，这是一种Android恶意软件表征和识别方法，它使用深度学习算法来解决当前对恶意软件检测的需求，使其在学习中变得更加自主，以更少的人为干预来解决问题。实验结果表明，与现有广泛使用的恶意软件检测方法相比，DroidDeepLearner方法取得了良好的性能。
With the proliferation of Android apps, encounters with malicious apps (malware) by mobile users are on the rise as vulnerabilities in the Android platform system are exploited by malware authors to access personal or sensitive information with ill intentions, often with financial gain in mind. To uphold security integrity and maintain user confidence, various approaches have been studied in the field of malware detection. As malware become more capable at hiding its malicious intent through the use of code obfuscation, it becomes imperative for malware detection techniques to keep up with the pace of malware changes. Currently, most of the existing malware detection approaches for Android platform use semantic pattern matching, which is highly effective but is limited to what the computers have encountered before. However, their performance degrades significantly when it comes to identifying malicious apps they have never tackled before. In this paper, we propose DroidDeepLearner, an Android malware characteri-zation and identification approach that uses deep learning algorithm to address the current need for malware detection to become more autonomous at learning to solve problems with less human intervention. Experimental results have shown that the DroidDeepLearner approach achieves good performance when compared to the existing widely used malware detection approaches.