Scholarlink

<p>The growth of Android devices has increased the chance of malware attacks, and conventional detection techniques are inadequate in dealing with obfuscation, zero-day attacks, and new families of malware. Strong detection frameworks are needed to identify informative features and classify sophisticated malware behaviour precisely. Introduce a hybrid Android malware classifier model integrating generative transformer-based N-tuple contrastive learning for feature representation with a new Quantum Support Red Piranha Vector Regression (QSRPVR) method for accurate classification. The transformer part is able to learn deep semantic patterns, and the N-tuple contrastive learning strengthens feature discriminability and generalizability. The QSRPVR classifier utilizes quantum kernel estimation and Red Piranha optimization for the best feature selection and accurate regression. The method combines static and dynamic analysis for end-to-end behavioural profiling. Experimental results on four benchmark datasets, Drebin, AndroZoo, AMD, and VirusShare, prove the better performance of the model. The method proved to be 99.8% accurate on Drebin and AndroZoo, and 99.78% on AMD and VirusShare datasets. The system proves to be highly robust to obfuscation attacks and exhibits generalizability across malware variants. This research introduces a strong and resilient Android malware detection system that blends deep generative learning with hybrid quantum optimization. The model improves detection strength, evasion robustness, and practicality in securing Android ecosystems.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Robust Android Malware Detection: Leveraging Generative Transformer-Based Feature Extraction and Hybrid Optimization Techniques

  • 期刊论文

  • 发表日期：

    2025年12月24日

查看全文
  • K. M. Veeresh,
  • Bhukya Ramesh Naik

摘要

The growth of Android devices has increased the chance of malware attacks, and conventional detection techniques are inadequate in dealing with obfuscation, zero-day attacks, and new families of malware. Strong detection frameworks are needed to identify informative features and classify sophisticated malware behaviour precisely. Introduce a hybrid Android malware classifier model integrating generative transformer-based N-tuple contrastive learning for feature representation with a new Quantum Support Red Piranha Vector Regression (QSRPVR) method for accurate classification. The transformer part is able to learn deep semantic patterns, and the N-tuple contrastive learning strengthens feature discriminability and generalizability. The QSRPVR classifier utilizes quantum kernel estimation and Red Piranha optimization for the best feature selection and accurate regression. The method combines static and dynamic analysis for end-to-end behavioural profiling. Experimental results on four benchmark datasets, Drebin, AndroZoo, AMD, and VirusShare, prove the better performance of the model. The method proved to be 99.8% accurate on Drebin and AndroZoo, and 99.78% on AMD and VirusShare datasets. The system proves to be highly robust to obfuscation attacks and exhibits generalizability across malware variants. This research introduces a strong and resilient Android malware detection system that blends deep generative learning with hybrid quantum optimization. The model improves detection strength, evasion robustness, and practicality in securing Android ecosystems.