Explainable Human Activity Recognition: A Unified Review of Concepts and Mechanisms

Opacity of AI poses real deployment barriers: Deep learning models have improved human activity recognition performance, but their "opaque" nature limits trust and real-world deployment in critical applications like healthcare monitoring and assistive living—suggesting that accuracy alone is insufficient for responsible AI systems.

Explainability is essential for high-stakes applications: The paper identifies healthcare monitoring, assistive living, and smart environments as key domains where HAR operates; in these contexts, explainability becomes critical for enabling human understanding and informed decision-making rather than blind reliance on AI predictions.

Multiple explanation approaches exist but lack standardization: The review establishes a "unified perspective" distinguishing conceptual dimensions of explainability from algorithmic mechanisms, addressing prior ambiguities—indicating that the field currently lacks consistent frameworks for what "explainability" means, which could hinder responsible deployment.

Complexity of real-world data requires tailored explanations: HAR systems must handle temporal sequences, multiple sensor types, and semantic meaning simultaneously; the paper emphasizes that explanation methods must address these specific complexities rather than applying generic XAI techniques, suggesting one-size-fits-all approaches are inadequate.

Evaluation practices and deployment readiness remain underdeveloped: The review highlights "key challenges in achieving reliable and deployable XAI-HAR" and notes limitations in current evaluation practices, suggesting that responsible AI for activity recognition requires better standards for assessing whether explanations actually support human trust and decision-making.