Comparative analysis of pre- and post-classification ensemble methods for android malware detection

Abstract

The influence of portable devices in our day-to-day activities is of a concern due to possibilities of a security breach. A large number of malwares are concealed inside Android apps which requires high-performance Android malware detection systems. To increase the performance, we have applied ensemble learning at feature selection level (pre-classification) and at prediction level (post-classification). The features extracted are the API classes and for generating the model, extreme learning machine (ELM) has been used. The filter feature selection methods employed are Chi-Square, OneR, and Relief. The experimental results on a corpus of 14762 Android apps show that ensemble learning is promising and results in high performance as compared to the individual classifier. We also present a comparison of the pre- and post-classification ensemble approaches for the Android malware detection problem.