Besi: behavior learning and tracking with wearable and in-home sensors - a dementia case-study: poster abstract

Sensing driven behavior modeling is vital in health applications. Recent advances in machine learning and sensing technologies accelerate such efforts. While wearables facilitate continuous sensing, they lack the computational resources for on-board heavy-weight signal processing and model-based prediction. Moreover, continuous transmission to a remote server drains much energy to achieve reasonable battery life for practical use. The BESI (Behavioral and Environmental Sensing and Intervention) system addresses these challenges to achieve continuous and real-time prediction-based tracking of human behavior. It employs a network of embedded nodes to ensure continuous connection with the wearables, and distributes the feature extraction and the model prediction tasks among these nodes and a local server to achieve real-time performance. In a dementia case-study, the BESI system is used for tracking agitated behavior in patients. It has been deployed in 12 residences of dementia patients, each for 30 days; and is planned for 10 more 60-day deployments. The system operation, behavior modeling method, and some preliminary result on tracking performance are presented here along with a discussion on future plan for platform optimization and model performance improvement.