A Systematic Literature Review of Android Malware Detection Using Static Analysis

Ya Pan, Xiuting Ge, Chunrong Fang, Yong FanIEEE Access 2020 · DOI

Introduction

Android currently powers over 80% of the world's smartphones, making it a prime target for malicious actors. Malware often operates covertly—stealing sensitive data or hijacking browsers—without user awareness. To combat this, researchers employ two primary strategies: dynamic analysis (executing apps in a sandbox) and static analysis (examining code without execution). While dynamic analysis is thorough, static analysis is significantly faster, capable of scanning thousands of apps pre-installation. However, the research landscape is fragmented. This paper systematically organizes 98 significant studies published between 2014 and 2020 to identify effective techniques and existing gaps.

Methodology

The authors conducted a Systematic Literature Review (SLR) following a rigorous protocol across seven major electronic databases, including IEEE Xplore and the ACM Digital Library. Starting with thousands of candidates, the search was narrowed down to 98 primary studies based on strict inclusion and exclusion criteria.

The systematic review process followed by the authors.

Classification of Static Analysis Techniques

To provide a structured overview, the reviewed studies were classified based on the "features" extracted from the applications. The authors identified four distinct archetypes of static analysis:

  1. Android Characteristic-based: Focuses on manifest-level data such as requested permissions, API calls, and hardware requirements. This category accounts for approximately 53% of studies due to its low computational cost.
  2. Opcode-based: Analyzes raw Dalvik bytecode as a sequence of text, capturing the "DNA" of the app. This method showed the highest mean accuracy (97.58%) among all reviewed techniques.
  3. Program Graph-based: Constructs complex maps of data flows or function calls, such as Control Flow Graphs (CFGs).
  4. Symbolic Execution-based: A high-precision method that replaces real data with symbols to explore every possible execution path the code could take.
Taxonomy of the reviewed studies based on analysis techniques and empirical evidence.

The Dataset Landscape

The review identifies a mix of curated "in-the-lab" benchmarks and "in-the-wild" live samples:

  • Curated Benchmarks: Drebin and Genome remain the most frequent baselines for comparative research.
  • Live Samples: Virusshare and Virustotal are the primary sources for evolving malware.
  • Benign Data: Researchers typically source "clean" apps from the Google Play Store or the China Mobile Application Market.
  • Machine Learning Integration: 82% of the studies incorporate machine learning, with Ensemble Learning (combining multiple models) being the most popular approach (22%).
  • Neural Networks: In direct comparisons, neural network models consistently outperform traditional statistical models, especially on large, complex datasets.
  • Standardized Tooling: Apktool has emerged as the industry standard for decompiling Android apps, surpassing alternatives like Androguard or Flowdroid.

Performance and Metrics

Across all categories, static analysis techniques achieved mean accuracies exceeding 88%. Accuracy remains the most common metric for success, followed closely by Precision and Recall. Opcode-based methods led the field, likely because they capture the actual logic of the code rather than just its metadata.

Limitations and Future Directions

Despite high performance, several challenges remain. The authors highlight a "comparison gap" caused by researchers using different versions of datasets and performance metrics, making direct model comparisons difficult. Furthermore, static analysis remains vulnerable to evasion tactics like code obfuscation and dynamic loading. The authors conclude that the next frontier lies in hybrid models that combine the efficiency of static analysis with the deep behavioral insights of dynamic analysis, supported by a unified evaluation platform.

Summary updated

BibTeX
@article{pan2020systematic,
  author = {Ya Pan and Xiuting Ge and Chunrong Fang and Yong Fan},
  title = {{A Systematic Literature Review of Android Malware Detection Using Static Analysis}},
  journal = {IEEE Access},
  year = {2020},
  doi = {10.1109/access.2020.3002842},
}
Figure