Android Malware Detection Through ML-Based Analysis Of APK Permissions

Abstract: The exponential growth of Android-based devices has resulted in a worrying rise in the spread of malware through mobile applications. The surge in Android malware highlights the crucial need for strong security measures. Machine learning, focusing on APK permission analysis, offers a promising solution to detect harmful apps and protect users from security threats and privacy breaches. The model classifies the APK files as benign or malicious based on the permissions used by the model. Our paper, primarily research-based, focuses on comparing various available options for detecting malware to identify the most suitable real-time solution. We propose a malware detection system that assesses an app's maliciousness by analyzing its permission usage. This study presents an innovative method for detecting Android malware, employing Support Vector Machines (SVM), as the machine learning model of choice after evaluating other models. In addition to comparing various models, we incorporated feature reduction techniques during the assessment process. After a comprehensive comparison of various parameters among different models, Support Vector Machines (SVM) emerged as the most suitable choice for our research. The feasibility of SVM was determined through measures such as ROC-AUC, recall, precision, accuracy, and F1-score.

Keywords: Machine learning, APK, Malware, Android, permissions.