Scholarlink

<p>The increasing proliferation of Android malware has heightened the need for robust detection strategies, particularly in dynamic environments. While future applications may require real-time and cross-platform solutions, current malware detection systems suffer from high computational overhead, low adaptability to new features, inefficiencies in detecting evolving and sophisticated malware, and generalization across diverse malware families; hence, inability to effectively detect new attack techniques. Future applications of Android malware detection demand real-time solutions to these issues that would ensure robust device-agnostic protection across numerous platforms. To mitigate these constraints, this paper therefore introduces an Adaptive Clustering Wavelet Few-Shot Learning (ADCWFSL) framework that integrates adaptive clustering, wavelet multi-resolution analysis, and few-shot learning for a potentially more accurate, adaptive, and scalable detectionapproach. This framework uses a two-tier approach. The first tier uses Subspace-Embedded Adaptive Dip-Based Enhanced Clustering Kernel (SEADECK), a clustering-based method for feature extraction and dimensionality reduction, which identifies high-quality seed features to reduce the complexity of computations. The second tier is Adaptive Few-Shot Wavelet Multi-Resolution (AFSWMR), which enables better malware classification through wavelet transformations and few-shot learning to introduce better generalization across diversely and evolutionarily changing samples of malware. The empirical findings indicate that the ADCWFSL framework surpasses current approaches across precision, recall, F1-score, and accuracy metrics, yielding substantial performance gains across the two datasets used in the evaluation.</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

Adaptive latent space dip clustering and few-shot wavelet learning for android malware detection

  • 期刊论文

  • 发表日期：

    2026年1月11日

查看全文
  • K. Selvi,
  • P. Vijayalakshmi,
  • B. Selvalakshmi,
  • G. Manikandan

摘要

The increasing proliferation of Android malware has heightened the need for robust detection strategies, particularly in dynamic environments. While future applications may require real-time and cross-platform solutions, current malware detection systems suffer from high computational overhead, low adaptability to new features, inefficiencies in detecting evolving and sophisticated malware, and generalization across diverse malware families; hence, inability to effectively detect new attack techniques. Future applications of Android malware detection demand real-time solutions to these issues that would ensure robust device-agnostic protection across numerous platforms. To mitigate these constraints, this paper therefore introduces an Adaptive Clustering Wavelet Few-Shot Learning (ADCWFSL) framework that integrates adaptive clustering, wavelet multi-resolution analysis, and few-shot learning for a potentially more accurate, adaptive, and scalable detectionapproach. This framework uses a two-tier approach. The first tier uses Subspace-Embedded Adaptive Dip-Based Enhanced Clustering Kernel (SEADECK), a clustering-based method for feature extraction and dimensionality reduction, which identifies high-quality seed features to reduce the complexity of computations. The second tier is Adaptive Few-Shot Wavelet Multi-Resolution (AFSWMR), which enables better malware classification through wavelet transformations and few-shot learning to introduce better generalization across diversely and evolutionarily changing samples of malware. The empirical findings indicate that the ADCWFSL framework surpasses current approaches across precision, recall, F1-score, and accuracy metrics, yielding substantial performance gains across the two datasets used in the evaluation.