Smart Home Residents’ Behavior Analysis
Primary author: Beiyu Lin
Faculty sponsor: Diane Cook
Primary college/unit: Voiland College of Engineering and Architecture
In 2030, 19 percent of the population in the United States will be aged 65 and older. In 2050, it will be 22 percent. With population growth and aging problems, we anticipate that there will be increasing healthcare needs of seniors for their physical and mental health problems. We want to design technology to help them live independently as long as possible at home and help them have a positive quality of life.
With decades of behavioral data from over 100 smart homes, we now can design new approaches to model human behavior from smart home sensors for extracting insights about our health. We design a new approach based on inverse reinforcement learning, which considers a house plan as a grid and each cell in the grid includes spatial-temporal features of a resident. For example, we design methods to study a resident’s in-home trajectory during the time s/he is healthy and then use deviations from this learned function to predict abnormal behaviors which may indicate potential health problems. Residents who make changes in their routine, such as sleeping in a living room recliner rather than a bed, are due to their health deterioration, such as increased breathing difficulties.
We are the first group to utilize inverse reinforcement learning to study indoor behavior patterns and its indication of health conditions. This model will help researchers having a greater understanding of human routine behavior and its variations that can transform how healthcare services are delivered to millions of homes.