Android Malware Detection Technique for IoT Devices Using Machine Learning

Abstract

Internet of things (IoT) based Android devices is revolutionizing our world with its problem-solving applications in different aspects of life such as education, agriculture, healthcare, sensing and remote monitoring, and so on. Android applications are working hand to hand nearly to make IoT dreams real. The number and scope of Android devices keep on increasing. It is assessed that there will be approximately 6.2 billion smartphone users by 2020/2021. Therefore, malicious program can affect the working of many devices shared in a network. The recent studies focus on malware detection for Android based devices but many researchers use various techniques and limited malware families for analysis. The goal of this study is to improve the Machine Learning algorithm performance through training dataset with features that has high importance score. The machine learning classification algorithm is implemented to extract the various features of Android malware using enhanced Decision tree classifier and implemented in three ML algorithms. The machine learning model is trained with datasets from three popular data sources, those are google play store, virusshare and Kaggle data. Moreover, Decision Tree, Naïve Bayes and SVM algorithms are compared to find better detection rate and hybrid analysis method is used for extracting more important features to provide classification for achieving high accuracy and robustness. The algorithms are implemented on the selected features with different running time and different numbers of apk and features, and the SVM classifier performed with better accuracy than others that is up to 97%. While the Decision tree and Naïve Bayes classifier performs less accurate. Therefore, according to this study the SVM algorithm is the better to Android malware detection.