From Code to Security: Machine Learning Approaches in Android Vulnerability Detection

Abstract

In today’s technology-driven society, an increasing number of individuals rely on mobile devices, leading to a surge in the availability of applications. Smartphone users constantly search for apps that meet their needs, resulting in a flood of options in the marketplace. However, there is a growing concern regarding the security of Android applications, as many have shortcomings in addressing critical security aspects. One reason behind this issue often lies in the lack of automated mechanisms during the design and development stages to identify, test, and rectify vulnerabilities in the source code. It is crucial to address these issues proactively rather than relying solely on updates and patches for already published apps. In response to this challenge, researchers have proposed machine learning techniques to enhance application security by detecting vulnerabilities and malicious code within source code. This systematic literature review delves into this domain by examining 85 carefully selected technical studies published between 2017 and 2024. It aims to shed light on the strengths, weaknesses, and practical applicability of these techniques, while also identifying areas for further improvement. Moreover, the growing focus on advanced approaches—such as Large Language Models (LLMs) and Explainable AI (XAI)—indicates a trend toward more transparent and context-aware vulnerability detection. By synthesizing key insights from the current literature, this review enhances our understanding of Android security approaches, identifies promising directions for future research, and ultimately contributes to the advancement of more secure mobile applications through machine learning-based vulnerability detection.