Scholarlink

The growing sophistication and prevalence of Android malware pose significant challenges to effective detection. Traditional methods often depend on extensive labeled datasets and struggle with the diversity of malware types. This study introduces an advanced Android malware detection framework that integrates self-supervised learning (SSL) with multiple machine learning models. SSL enables the extraction of robust feature vectors, allowing the model to identify meaningful patterns in data without heavy reliance on labeled samples. These features are utilized by classifiers, including random forest, logistic regression, SVM, gradient boosting, and K-nearest neighbors. Experimental results confirm that the hybrid SSL approach significantly improves detection accuracy, with random forest achieving the highest accuracy and F1-score of 90%, compared to 85% accuracy and 84% F1-score without SSL. The proposed methodology demonstrates the potential of SSL-driven feature extraction combined with machine learning models to address the limitations of traditional approaches.

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

Advanced Android Malware Detection with Self-supervised Learning and Multi-model Classification

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Venkata Subbaiah Desanamukula,
  • P. Sujatha,
  • Gottala Parameswara Kumar,
  • Vunnava Dinesh Babu,
  • Kusuma Polanki,
  • A. Lakshmana Rao

摘要

The growing sophistication and prevalence of Android malware pose significant challenges to effective detection. Traditional methods often depend on extensive labeled datasets and struggle with the diversity of malware types. This study introduces an advanced Android malware detection framework that integrates self-supervised learning (SSL) with multiple machine learning models. SSL enables the extraction of robust feature vectors, allowing the model to identify meaningful patterns in data without heavy reliance on labeled samples. These features are utilized by classifiers, including random forest, logistic regression, SVM, gradient boosting, and K-nearest neighbors. Experimental results confirm that the hybrid SSL approach significantly improves detection accuracy, with random forest achieving the highest accuracy and F1-score of 90%, compared to 85% accuracy and 84% F1-score without SSL. The proposed methodology demonstrates the potential of SSL-driven feature extraction combined with machine learning models to address the limitations of traditional approaches.