Introduction Caregiver burden associated with dementia-related agitation is one of the most common reasons community-dwelling persons with dementia (PWD) transition to a care facility. Agitation in dementia can be defined as “inappropriate verbal, vocal, or motor activity that is not explained by needs or confusion per se” and is “consistent with emotional distress”. Agitation can be physically non-aggressive, aggressive, or verbally agitated behavior. Up to 90% of PWDs experience agitation and it is one of the principal factors for institutionalization. Agitation can be unpredictable, remain undetected in the early stages, and quickly escalate. Caregivers of PWD report high levels of emotional, physical, and financial stress and are susceptible to disease and health complications. BESI is an interactive, cybersociophysical system comprising of in-home and body-worn sensors, a caregiver tablet interface, online modeling, and caregiver notifications built to help detect the environmental triggers and early signs of agitation in PWD. Methods This is a mixed method, prospective study using descriptive, qualitative, and quantitative measures of caregiver-PWD dyads. Baseline psychosocial and behavioral status for both members of the dyad were assessed using validated assessment tools. Phase 1 (System Verification) - controlled settings (laboratory and homes of 2 healthy volunteers) and study settings (homes of 2 study dyads) & usability testing Phase 2 (System Effectiveness) - homes of 10 study dyads for 30 days each Home Visits First: provide detailed description of study, obtain informed consent, and document demographic and medical information. Second: complete assessment battery and BESI system installation. Results Phase 1 The tablet application usability was assessed and revised showing it to be a viable tool. We observed and documented caregiver emotional-state evolution over the day in relation to agitation episodes, the correlation between agitation events and accelerometer Teager energy magnitudes, and the variability in environmental parameters across rooms temporally and during agitation episodes. Audio levels did not yield enough information for more complex audio processing such as verbal agitation detection. We concluded that additional model training and evaluation was needed for physical agitation detection and PWD room localization. Phase 2 An upgraded audio data collection system was implemented to provide more complex audio processing. Comparing this with the verbal agitation episode reports on the tablet application, this method produced a 76.7% accuracy in detecting verbal agitation events. Multiple models were explored for physical agitation detection, with long short-term memory (LSTM) cell based recurrent neural network providing the highest performance. Decision rules was used to structure covariance factor retention for each dyad to feed into artificial neural network models. Doorway-based sensors and Markov models with room transition probabilities provided ∼90% room-level tracking accuracy of the PWD. Conclusions Phases 1 & 2 results facilitated targeted changes in BESI, thus improving its overall usability for the final phase of the study. This research was funded by This study is funded by the National Science Foundation Smart and Connected Health Program. [Grant Number 1418622]