Android Malware Prediction Using Extreme Learning Machine with Different Kernel Functions

Abstract

Android is currently the most popular smartphone platform which occupied 88% of global sale by the end of 2nd quarter 2018. With the popularity of these applications, it is also inviting cybercriminals to develop malware application for accessing important information from smartphones. The major objective of cybercriminals to develop Malware apps or Malicious apps to threaten the organization privacy data, user privacy data, and device integrity. Early identification of such malware apps can help the android user to save private data and device integrity. In this study, features extracted from intermediate code representations obtained using decompilation of APK file are used for providing requisite input data to develop the models for predicting android malware applications. These models are trained using extreme learning with multiple kernel functions ans also compared with the model trained using most frequently used classifiers like linear regression, decision tree, polynomial regression, and logistic regression. This paper also focuses on the effectiveness of data sampling techniques for balancing data and feature selection methods for selecting right sets of significant uncorrelated metrics. The high-value of accuracy and AUC confirm the predicting capability of data sampling, sets of metrics, and training algorithms to malware and normal applications.