Privacy-Preserving Explainable AIoT Application via SHAP Entropy Regularization

Researchers developed a privacy-preserving method using SHAP entropy regularization to protect sensitive user data in explainable AI systems for smart home IoT applications. The approach reduces privacy leakage while maintaining model accuracy and explanation quality.

• →Current explainable AI methods like SHAP and LIME can inadvertently expose sensitive user attributes and behavioral patterns in smart home environments.
• →The proposed SHAP entropy regularization method promotes uniform distribution of feature contributions to reduce privacy risks.
• →Researchers developed privacy attack methods to test their system's effectiveness against explanation-based data inference.
• →Experimental results show significant privacy leakage reduction while maintaining high predictive accuracy and explanation fidelity.
• →The work addresses growing regulatory compliance needs for transparent yet privacy-preserving AI systems in IoT applications.