Project Abstract Experiences of loneliness are prevalent among older adults. Loneliness is a painful and pernicious state occurring when there is a perceived discrepancy between one’s optimal levels of social interactions and actual social relationships. Given that loneliness has been associated with negative health outcomes, detection measures are imperative to identify both lonely and at-risk individuals early enough to intervene before adverse health outcomes occur. However, the assessment of loneliness is challenging due to the stigma of being labeled as lonely and the lack of integration of loneliness assessments into primary care or community-based services. To tackle the problem, we will use smart speakers and passive sensing data to automatically assess older adults’ level of loneliness as correlates of self-report scores on the UCLA Loneliness Scale. Our goal in this project is to develop, deploy, and validate a smart speaker-based system for in-home loneliness assessment that integrates Internet of Things (IoT) devices to gather both speech (acoustic and prosodic features) and behavioral data (e.g., smart speaker use patterns, in-home mobility, sleep) from older adults. We will enroll 70 individuals (age 65+) living alone in the community to collect digital biomarker data continuously for 3 months. The specific aims are to: (1) Develop an innovative remote loneliness assessment system that allows passive and unobtrusive capture of speech and behavioral data in the home setting; (2) Using a train- test approach, develop and evaluate the performance of novel multi-class machine learning (ML) algorithms— semi-supervised type of Generative Adversarial Networks (SGANs) — to estimate older adults’ loneliness scores. The UCLA Loneliness Scale will be completed by participants every two weeks for 3 months to collect ground truth data; and (3) Identify potential implementation barriers to the effective use of the system among older adults (n=30). In particular, we will assess potential privacy and security concerns, social influence, cultural values, and the level of personification of virtual agents that could influence the adoption and use of the system. The novel ML models will allow us to identify not only “already” lonely individuals but also “at-risk” individuals. The proposed research seeks to provide preliminary evidence that digital biomarkers from smart speakers and IoT devices can be used for automatic loneliness assessment in the community. Results from this study will support our long-term goal of implementing the system longitudinally as a platform for real-time loneliness assessment and detection. This work will provide an opportunity to examine associations between digital speech and behavioral data and self-report measures of loneliness to identify modifiable risk factors that will inform the design of interventions to prevent adverse health impacts of loneliness and improve social well- being among the older adult population.