Exploring Deep Reinforcement Learning for Android Malware Detection

EasyChair Preprint 6594

Abstract

In today's world of technology, we are surrounded by electronic gadgets and connected through the internet. The risk of privacy invasion is at an all-time high. Cybersecurity plays an important role in preventing and precluding these threats and provides a safeguard against these attacks. In this paper, we have focused on Android Malware detection, since its causes have been seen skyrocketing due to easy access to a host device, which makes it susceptible to attacks and data breach. Moreover, Android is open-source, due to which it exposes the application to foreign attacks. Drebin dataset - the most extensive dataset for android malware detection, which is extracted from 15036 application files, was used for analyzes. The feature extraction was performed using Random Forest Classifier and Extra Trees Classifier, top 15 features were selected, and the accuracies were compared. The key focus of our work is to map the Android malware detection problem into the Markov Decision Process which is the mathematical foundation of the Reinforcement learning algorithm. We have implemented Q-Learning, a Deep Reinforcement Learning technique, for the classification of android malware on the Drebin dataset. The accuracy of RL (94.30%) was compared with the performance of other malware detection techniques.

Keyphrases: Android Malware Detection, Extra Trees Classifier, Markov Decision Process, Q-learning, Random Forest Classifier, Reinforcement Learning

@booklet{EasyChair:6594,
  author    = {Simran Gill and Surabhi Gogte and Chahak Sharma and Prathmesh Pathwar and Viraj Desai and Ashutossh Gupta and Aamod Vyas and O.P. Vyas},
  title     = {Exploring Deep Reinforcement Learning for Android Malware Detection},
  howpublished = {EasyChair Preprint 6594},
  year      = {EasyChair, 2021}}