Malware Diagnosis : Dynamically Detecting Android Malware with Weighted Permissions Using Deep Learning




Journal Title

Journal ISSN

Volume Title



Android mobile phone has rapidly become popular and irreplaceable. The open-source Android platform allows developers to innovate the Android market in various ways, but also raises significant issues with various malicious apps, such as device malfunction, personal information leak, or financial loss. Yet, it is difficult to detect malicious apps by a human or obtain explicit information about suspicious apps. To solve the problem, many studies have come up with some frameworks. However, many frameworks have constraints such as only running on PC and manual data processing. In this thesis, we propose the Malware Diagnosis framework for deep learning-based malware detection using weighted permission. It is designed to be more practical to use with better performance in detecting malware apps. To increase the accuracy of the framework, we apply a ranking-based approach to permissions to generate weights that are derived from the ranking based on the number of permission used from malware and benign apps. As a tool, we develop MD (Malware Diagnosis) Assistant, an Android app that performs automated data extraction from installed apps and provides a prediction rate by running a deep learning model on an Android device. We then present experimental observations that show the effectiveness of our framework on detecting malware apps.



Android (Electronic resource), Malware (Computer software), Machine learning