# Identification of Significant Permissions for Efficient Android Malware Detection

**Authors:** Hemant Rathore, Sanjay K. Sahay, Ritvik Rajvanshi, Mohit Sewak  
**arXiv:** [2103.00643](https://arxiv.org/abs/2103.00643)  
**Date:** 2021-02-28  
**Summary page:** https://www.summarizepaper.com/en/paper/2103.00643v1/

## Key points

- Malware on Android OS for smartphones is a significant concern due to the 3Vs of volume, velocity, and variety contributing to their growth
- One in every five business/industry mobile applications leaks sensitive personal data
- Traditional signature/heuristic-based malware detection systems are inadequate in dealing with current malware challenges on Android OS
- Researchers have turned to machine learning and deep learning-based malware detection systems
- A study titled "Identification of Significant Permissions for Efficient Android Malware Detection" conducted a comprehensive feature analysis to identify crucial Android permissions necessary for efficient malware detection using machine learning and deep neural networks
- The authors derived a set of 16 permissions (8% of the total set) to construct a malware detection engine that requires less training and testing time without compromising model accuracy significantly
- The random forest classifier-based Android malware detection model achieves an impressive area under curve score of 97.7%, surpassing current state-of-the-art systems' performance
- Deep neural networks attain comparable accuracy levels as baseline results but come at a massive computational penalty
- This study's findings provide valuable insights into identifying critical Android permissions necessary for efficient malware detection using machine learning and deep neural networks while highlighting the limitations of traditional signature/heuristic-based approaches in addressing current malware challenges on Android OS for smartphones.

## Summary

The proliferation of malware on Android OS for smartphones has become a significant concern, with the 3Vs of volume, velocity, and variety contributing to their growth. A recent report reveals that one in every five business/industry mobile applications leaks sensitive personal data. Traditional signature/heuristic-based malware detection systems are inadequate in dealing with current malware challenges, posing a threat to the Android ecosystem. As such, researchers have turned to machine learning and deep learning-based malware detection systems. In this study titled "Identification of Significant Permissions for Efficient Android Malware Detection," Hemant Rathore, Sanjay K. Sahay, Ritvik Rajvanshi, and Mohit Sewak conducted a comprehensive feature analysis to identify the crucial Android permissions necessary for efficient malware detection using machine learning and deep neural networks. Using variance threshold, auto-encoders, and principal component analysis techniques, the authors derived a set of 16 permissions (8% of the total set) to construct a malware detection engine that requires less training and testing time without compromising model accuracy significantly. The experimental results show that the random forest classifier-based Android malware detection model is well-balanced and achieves an impressive area under curve score of 97.7%, surpassing current state-of-the-art systems' performance. While deep neural networks attain comparable accuracy levels as baseline results, they come at a massive computational penalty. Overall, this study's findings provide valuable insights into identifying critical Android permissions necessary for efficient malware detection using machine learning and deep neural networks while highlighting the limitations of traditional signature/heuristic-based approaches in addressing current malware challenges on Android OS for smartphones.