Agitation is one of the most common and distressing behavioral and psychological symptoms of dementia, yet predicting its onset remains challenging.

In this study, we investigated whether passive in-home passive infrared (PIR) sensor data could predict next-day agitation using the Technology Integrated Health Management (TIHM) dataset.

We analyzed 56 community-dwelling adults with dementia, comprising 2,610 patient-days and 114 agitation events.

Behavioral features extracted from PIR sensors were evaluated using leave-one-patient-out cross-validation.

The best-performing model achieved an AUROC of 0.848 and a precision-recall AUC 6.1 times above the chance level.

Activity volume, evening activity, and self-loop ratio were identified as the strongest behavioral predictors.

These findings suggest that passive in-home sensing may provide a scalable, low-burden approach for early detection of agitation and support proactive dementia care.