A Deep Learning Approach for Android Malware Detection Using Mixed Bytecode Images and Attention-Based ResNet

Authors

  • Naeema Nazar Author
  • AlexBenny Author

Keywords:

bytecode, ResNet, convolutional, generalization

Abstract

Detecting Android malware becomes crucial at a time when apps are becoming more and more prevalent and controlling almost everything we do on our smartphones. In order to overcome these limitations in signature-based detection and combat new and constantly changing malware varieties, our paper suggests a deep learning approach. Using this method, an Android application's bytecode is transformed into a collection of graphics that depict the program's dynamic (behavior and structure) aspects. These pictures will serve as the foundation for successfully using sophisticated image classification techniques to identify viruses. An attention-amplified ResNet, one of the cutting-edge convolutional neural network architectures that may be used to image recognition, is at the center of this system. The model's ability to differentiate between benign and malicious programs with greater focus weights is improved by the inclusion of the attention mechanism, which enables the model to pay more (less) attention to specific areas in the associated bytecode pictures. By combining mixed bytecode images—that is, original and augmented versions—the method enhances a model's adversarial resilience for withstanding novel evasion methods and raises its generalization qualities to other malware families. Experiments conducted on a large dataset of Android apps demonstrate that recall, accuracy, and precision all outperformed the malware detection algorithms currently in use. It turns out that enhancing detection effectiveness, particularly on highly obfuscated malware, is largely dependent on the attention mechanism. Additionally, using mixed bytecode pictures lowers false positives and boosts accuracy to a point where these models may be used in real-world scenarios. Our research supports the use of deep learning and image-based analysis for Android malware detection. We will be able to create new, more sophisticated security solutions that are more appropriate for the dynamic mobile threat landscape of today thanks to our attention-based ResNet architecture.

Downloads

Published

2026-06-08