HYBRIDROID: Detecting Malicious Applications on Linux Powered Smartphone’s by Hierarchical Machine Learning Algorithms.

Abstract

From the last two decades mobile phones are a very important part of our lives and Android is known to be the most popular working framework for mobile phones. Android is having a quickly expanding scenario that has pulled in the intruders for creating malwares. Android permits downloading and installation of applications from android market and third party websites. This offers malware engineers a chance to put repackaged malicious code in the legitimate applications. Many malware detection techniques are proposed and various discovery frameworks have been created which utilizes static and dynamic investigation techniques. However, the existing models are not sufficient for accurately detecting malicious applications in the Android Smartphones. Many Malware Detection Techniques are good in specific areas like selection of maximum and informative features, Accuracy of Detection, on device analysis and Minimum Performance Overhead. To increase the ac- curacy and efficiency of Malware detection we proposed multilayer models using fusion of algorithms in the field of machine learning. The proposed model is a complete solution by considering all best possible algorithms in each area maintaining efficiency and accuracy in the Detection System. The research contributes in multiple layers. First, In Feature extraction Maximum features are considered including user level and Kernel Level. At Feature Selection most informative Features selected for accurate results. A novel fusion approach is proposed by using various machine learning algorithms for classification of applications according to the behavior. This provides high efficiency and accuracy by combining Static and dynamic analysis of Malware detection. Experimental results have shown high accuracy with great efficiency in terms of memory as well as Power.