Scholarlink

<p>The rapid growth of Android malware calls for accurate and efficient detection techniques. Function call graph (FCG) based static analysis has shown strong potential for modeling application behaviors, yet existing approaches commonly face two challenges: (i) when simplifying large-scale call graphs to reduce overhead, they often fail to adequately preserve the critical contextual information around sensitive APIs; and (ii) continuous Android API evolution tends to induce method-level feature drift and render sensitive API sets outdated, leading to feature mismatch and degraded generalization in cross-year scenarios. These issues jointly weaken a model’s ability to capture the semantics of malicious behaviors, with the performance drop being especially evident under cross-year evaluation. To address this, we propose FUADroid, a static malware detection method that fuses structural and statistical-semantic views. FUADroid first constructs an updatable sensitive API set by combining knowledge priors from Android official documentation with a <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\chi ^2\)</EquationSource> </InlineEquation> statistical test. It then builds a Sensitive System Function Call Graph (SSFCG) by retaining n-hop contextual subgraphs around sensitive APIs while aggregating peripheral structures into class-level supernodes, thereby balancing semantic fidelity and computational efficiency. Finally, FUADroid performs soft-voting fusion between a GCN-based structural classifier on SSFCG and a Random Forest classifier on class-level statistical semantic features to enable complementary modeling. Experimental results on multiple datasets demonstrate that FUADroid achieves strong detection performance and exhibits improved robustness under cross-year evaluation settings.</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

FUADroid: android malware detection with multi-view API feature fusion using machine learning

  • 期刊论文

  • 发表日期：

    2026年7月22日

查看全文
  • Jiyun Yang,
  • Fan Mei,
  • Zhengdong Wan,
  • Xintong Cai,
  • Pei Ran

摘要

The rapid growth of Android malware calls for accurate and efficient detection techniques. Function call graph (FCG) based static analysis has shown strong potential for modeling application behaviors, yet existing approaches commonly face two challenges: (i) when simplifying large-scale call graphs to reduce overhead, they often fail to adequately preserve the critical contextual information around sensitive APIs; and (ii) continuous Android API evolution tends to induce method-level feature drift and render sensitive API sets outdated, leading to feature mismatch and degraded generalization in cross-year scenarios. These issues jointly weaken a model’s ability to capture the semantics of malicious behaviors, with the performance drop being especially evident under cross-year evaluation. To address this, we propose FUADroid, a static malware detection method that fuses structural and statistical-semantic views. FUADroid first constructs an updatable sensitive API set by combining knowledge priors from Android official documentation with a \(\chi ^2\) statistical test. It then builds a Sensitive System Function Call Graph (SSFCG) by retaining n-hop contextual subgraphs around sensitive APIs while aggregating peripheral structures into class-level supernodes, thereby balancing semantic fidelity and computational efficiency. Finally, FUADroid performs soft-voting fusion between a GCN-based structural classifier on SSFCG and a Random Forest classifier on class-level statistical semantic features to enable complementary modeling. Experimental results on multiple datasets demonstrate that FUADroid achieves strong detection performance and exhibits improved robustness under cross-year evaluation settings.