Explainable Activity Recognition for Smart Home Systems

Users strongly prefer explanations over simple labels: In 83% of sampled scenarios, non-expert users preferred natural language explanations over a simple activity label when monitoring remote care recipients, demonstrating clear demand for transparency in smart home decision-making.

XAI methods can reliably explain smart home decisions: The SHAP explainability technique achieved a 92% success rate in generating sensible explanations for activity recognition, suggesting that existing AI explanation methods can be effectively applied to make smart home systems more trustworthy.

Explainability is critical for safety and trust: Because smart home systems make automated decisions that affect vulnerable populations (like elderly care recipients), users need to understand why the system took an action—not just what action it took—to catch errors and maintain confidence in the technology.

Different explanation methods may work better in different contexts: The paper indicates that explanation effectiveness varies depending on the scenario (synchronous vs. asynchronous monitoring), suggesting that responsible AI implementation requires tailoring explanations to specific use cases rather than applying one-size-fits-all solutions.

Expert validation alone is insufficient: The research used both ML experts and non-expert users to evaluate explanations, highlighting that responsible AI requires testing with actual end-users, not just technical specialists, to ensure explanations are truly understandable and useful.