An Efficient Android Malware Prediction Using Ensemble machine learning algorithms

Abstract

Malwares are designed to disrupt, disable or take control of a computer system. Android malware specially targets Android OS through leakage of confidential information and crashing the system. Several attempts have been made to detect Android malware. However, those works are unable to detect malware automatically and most of them are signature based which cannot detect new variants of malware. In our work, we have explored different algorithms to obtain the best algorithm for malware prediction and to obtain the best set of features that will help us in predicting malware efficiently. From our analysis, we have seen that ensemble methods are better than traditional machine leaning algorithms for predicting malware. We have reduced the number of features from 215 to 100 achieving an accuracy of 99.5% using Light GBM. In addition, we have obtained an accuracy of 99.1% using Random Forest having only 55 features.