Android is the world's most popular and widely used operating system for mobile smartphones today. One of the reasons for this popularity is the free third-party applications that are downloaded and installed and provide various types of benefits to the user. Unfortunately, this flexibility of installing any application created by third parties has also led to an endless stream of constantly evolving malware applications that are intended to cause harm to the user in many ways. In this project, different approaches for tackling the problem of Android malware detection are presented and demonstrated. The data analytics of a real-time detection system is developed. The detection system can be used to scan through installed applications to identify potentially harmful ones so that they can be uninstalled. This is achieved through machine learning models. The effectiveness of the models using two different types of features, namely permissions and signatures, is explored. Exploratory data analysis and feature engineering are first implemented on each dataset to reduce a large number of features available. Then, different data mining supervised classification models are used to classify whether a given app is malware or benign. The performance metrics of different models are then compared to identify the technique that offers the best results for this purpose of malware detection. It is observed in the end that the signatures-based approach is more effective than the permissions-based approach. The kNN classifier and Random Forest classifier are both equally effective in terms of the classification models.
Professional Studies (MS)
Department, Program, or Center
Graduate Programs & Research (Dubai)
Alhebsi, Mohamed Salem, "Android Malware Detection using Machine Learning Techniques" (2022). Thesis. Rochester Institute of Technology. Accessed from