Large Language Models for Human Activity Recognition in Smart-Home

The sensor-based recognition of Activities of Daily Living (ADLs) in smart-home environments enables applications in healthcare, including monitoring behavioral changes in older adults to detect cognitive decline. However, traditional deep learning approaches face two key challenges: reliance on large labeled datasets and lack of transparency. This research explores the potential of Large Language Models (LLMs) to address these challenges. To mitigate data scarcity, we tested LLMs as substitutes for knowledge models in neuro-symbolic approaches and evaluated their ability to recognize activities from sensor data. For explainability, we developed methods to assess LLM-generated explanations and investigated whether LLMs, in addition to classifying activities, can generate effective natural language explanations. Overall, our findings suggest that LLMs can enhance ADLs recognition by alleviating data and explainability limitations, though challenges remain, which are discussed in this work.