Android Malware Prediction by Permission Analysis and Data Mining

Abstract

In recent years, smartphones have brought people’s lives to a new high level. Smartphone applications, or Apps, are accelerating the process with many more functions getting developed, such as browsing the Internet, making payments, taking photos and share. However, the "Apps" are bringing potential vulnerability when they access private information from the phones, and mobile security has never been so much focused on like today. In this paper, we presented a novel An-droid Permission based malware detection technique. We first gather a huge set of both malware and benign Apps through web clawer and develop a tool to decompile Apps to source code and manifest files automatically. Then permissions with other information are extracted for each App, making up to a raw data set. Afterward, we apply data cleaning, dimension reduction and statical analysis to the raw data set. We find that the distribution of permissions for Apps shares a differ-ence between malware dataset and benign dataset. Finally, we take advantage of machine learning algorithms, including Logistic Regression Model, Tree Model with Ensemble techniques, Neural Network and finally an ensemble model to find patterns and more valuable information. Other models are also discussed. Extended experiments using these various machine learning models are conducted in the end. From the results, we can see that our method generates a good accuracy, F-score and overall performance of malicious App prediction.