A BPSO and Deep Learning Based Hybrid Approach for Android Feature Selection and Malware Detection

With the rapid expansion of smartphone technology, smartphone has become the most influential tool to access the Internet and avail myriad services at a click. On the contrary, app vulnerabilities and malware have become the main threats to the security of Android devices. In general, large datasets are used for malware detection, which may contain several inappropriate and redundant attributes, which can increase the misclassification rate and computational time. As a result, we require a system that can eliminate unwanted features and improve detection accuracy. The authors, here, present a Binary Particle Swarm Optimization (BPSO) and a Deep learning-based highly efficient hybrid approach that helps to address this complex issue of critical relevance. The BPSO method extracts the best features from the search space of API calls, intentions, and command signatures, while Deep Neural Networks (DNNs) assess subsets using data insights. The proposed approach has been experimentally verified with the standard android ‘DREBIN’ dataset. This hybrid approach achieves a significant detection precision of 97.92%, a Recall rate of 96.35 %, an accuracy of 94.92 %, and an F1- score of 92.39% with 72 optimal features.

I. Introduction

Smartphones have become an essential part of people's lives due to several applications, like e-shopping, internet banking, internet surfing, social networking, etc. According to Statista, the number of smartphone users worldwide is about 3.8 billion which is expected to increase by several hundred million in the coming few years [1]. Android operating system holds 73% market share as compared to other platforms, and has maintained popularity in the global market as of June 2021 [2]. According to the report of the first quarter of 2021, Google Play Store is the largest app store in the world and has 2.96 million apps available [3].