ENHANCING ANDROID SECURITY: MACHINE LEARNING APPROACHES FOR APP PERMISSIONS AND MALWARE DETECTION

Abstract

The prevalence of Android devices has made the platform a major player in the mobile market. However, this widespread adoption also brings significant challenges when de tecting and preventing Android malware. This thesis aims to address these challenges by utilizing machine learning techniques. The first aspect of this thesis focuses on giving users greater control over their app permissions through an innovative Android applica tion. This application uses machine learning models to analyze the permissions requested by various apps and provides users with informed recommendations based on the safety ratings of these permissions. By leveraging usage frequency data from specific app cat egories on the Google Play Store, the model offers users a comprehensive tool to make educated decisions regarding app permissions, thus contributing to a safer and more secure mobile app ecosystem. The second aspect of this thesis introduces the GARB (Gradient Boosting classifier, Ad aBoost, Random Forest and Bagging classifier with a decision tree) Model, a novel ensem ble approach for detecting malware in Android packages. The GARB Model combines multiple base classifier algorithms, including Gradient Boosting, AdaBoost, Random For est, and Bagging Classifier with Decision Tree, through a weighted averaging method of stacking. This ensemble approach exhibits superior performance, achieving an accuracy of 82.38% in distinguishing between malware and benign Android packages. Moreover, the GARB Model outperforms individual classifiers in various criteria, highlighting its efficacy and reliability for prediction tasks related to malware detection. Together, these contributions offer comprehensive solutions for enhancing mobile app se curity and user privacy within the Android ecosystem. By leveraging machine learning techniques, this thesis aims to mitigate the growing threat of Android malware and em power users with the tools and knowledge necessary to navigate the increasingly complex landscape of mobile app permissions and security risks.