A machine learning approach for detection of android malware based on hybrid analysis

Android is an open source Smartphone OS developed by Google. Android is highly targeted
by malware apps in recent years. We suggest a hybrid approach which combines the static
and dynamic methods for malware detection in Android. In the static analysis, we extracted
the information from source code of android application by reverse engineering the
application and in dynamic analysis we have traced out the system calls generated by the
application by strace utility. This information is further used as the features of a machine
learning classifier. We used machine learning classifiers such as Decision Tree, KNN,
Logistic regression, Naive bayes, SVC and random forest. With Decision tree classifier we
got 58.33% accuracy, whereas KNN, Logistic Regression and Support Vector Classifier have
given an accuracy of 66.66%. Random forest and Naive bayes have given a top accuracy of
75%.

MSC CS 2016-2018 INT Dr. Tony Thomas