Scholarlink

Android malware poses an increasing threat to individual security. Despite the availability of numerous detection methods, their effectiveness and transparency remain significant concerns. Machine Learning (ML) and Deep Learning (DL) models are effective but often lack clear explanations for their predictions. This paper presents an explainable approach using Deep Neural Networks (DNN) in conjunction with Explainable AI (XAI) techniques, including SHAP (Shapley Additive Explanations) and LIME (Local Interpretable Model-agnostic Explanations). By integrating these methods, the approach provides deeper insights into the decision-making process of the model. SHAP and LIME effectively identify the key features that influence classification, enhancing the model’s transparency and trustworthiness. The dataset used for the experiment contains 215 features extracted from 1,260 malware and 2,539 benign samples. The proposed model focuses on static features and achieves an impressive accuracy of 99.95%. Through careful analysis, the model’s performance is evaluated not only on its predictive accuracy but also on the interpretability of its decisions. A set of sensitive features that impact the detection process based on the XAI interpretation are identified and presented. These features enhance the ability to fine-tune the model and strengthen cybersecurity measures for Android platforms.

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

Enhancing Transparency in Android Malware Detection Using Explainable SHAP-LIME with Deep Learning Technique

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Diptimayee Sahu,
  • Satya Narayan Tripathy,
  • M. Soujanya

摘要

Android malware poses an increasing threat to individual security. Despite the availability of numerous detection methods, their effectiveness and transparency remain significant concerns. Machine Learning (ML) and Deep Learning (DL) models are effective but often lack clear explanations for their predictions. This paper presents an explainable approach using Deep Neural Networks (DNN) in conjunction with Explainable AI (XAI) techniques, including SHAP (Shapley Additive Explanations) and LIME (Local Interpretable Model-agnostic Explanations). By integrating these methods, the approach provides deeper insights into the decision-making process of the model. SHAP and LIME effectively identify the key features that influence classification, enhancing the model’s transparency and trustworthiness. The dataset used for the experiment contains 215 features extracted from 1,260 malware and 2,539 benign samples. The proposed model focuses on static features and achieves an impressive accuracy of 99.95%. Through careful analysis, the model’s performance is evaluated not only on its predictive accuracy but also on the interpretability of its decisions. A set of sensitive features that impact the detection process based on the XAI interpretation are identified and presented. These features enhance the ability to fine-tune the model and strengthen cybersecurity measures for Android platforms.