Perbdroid: Effective malware detection model developed using machine learning classification techniques

Abstract

This chapter introduces PerbDroid—a framework to detect malware from Android smartphones. To address the issues of malware detection through a broad set of apps, researchers have recently started to identify the features which helps to detect malware from apps. The proposed framework is based on features selection techniques which help us to develop a useful model for malware detection. We collected a data set of 2,00,000 Android apps from distinct sources and extracted permissions and API calls from them (consider as features in this study). Further, features are selected by using six different feature ranking approaches (i.e., Gain Ratio, OneR feature evaluation, Chi-squared test, Information gain feature evaluation, Principal Component Analysis (PCA) and Logistic regression analysis) to develop the model for malware detection. We evaluated several machine learning algorithms and feature selection methods in identifying the combination that gives the foremost performance to detect malware from real-world apps. Empirical outcomes illustrate that the proposed framework is useful to detect malware from smartphones mainly and in particularly from Android.