An Intelligent Stacked Autoencoder Framework for Android Malware Detection: Static Analysis Empowering Cybersecurity
Android has become the most widely utilized mobile platform worldwide, making it a primary target for malware authors. Such threats pose significant risks to user privacy and sensitive data, while the growing sophistication of Android malware has diminished the effectiveness of traditional signature-based detection methods. To address these challenges, this study proposes a hierarchical static analysis framework that leverages sparse stacked autoencoders to enhance Android malware detection. Experiments were performed on the TUANDROMD dataset, comprising 4,465 Android applications, including 3,565 malware samples and 899 benign apps. The proposed framework consists of three stages: (i) selection of discriminative features, (ii) hierarchical feature extraction using sparse autoencoders, and (iii) final classification with stacked autoencoders followed by fine-tuning. The system achieved 98.9% accuracy, with Precision = 98.9%, Recall = 99.7%, and F1-score = 98.3%, including a notably high recall of 99.7% which minimizes false negatives. Comparative assessment confirmed superior performance over conventional machine learning and clustering approaches. Beyond outperforming traditional methods, the framework demonstrated scalability, resilience against obfuscation, as well as strong potential for deployment in large-scale mobile security systems. These findings establish TUANDROMD as a benchmark dataset for future deep learning research and highlight stacked autoencoders as a promising solution for advancing Android malware detection.
Authors
- Reza Ahsan (ORCID: https://orcid.org/0000-0003-0718-0844)
- Marziyh Khavari-Estahbanati
- Mostafa Ghobai-Arani
Institutions
- Islamic Azad University, Tehran (IR)
Publication Details
- Journal
- Research Square
- Published
- 2026-11-20
- DOI
- https://doi.org/10.21203/rs.3.rs-8908567/v1
- Primary Topic
- Advanced Malware Detection Techniques
- Type
- preprint