Towards The Use of Smart Home Sensor Networks to Generate Predictive Activity Models

There are many use cases in the areas of cognition studies, physical therapy, and other medical related fields that stand to benefit from the ability to study the activities of individuals at home instead of a clinical environment. By monitoring their daily movements, various behavioral models can be generated that can aid in the early detection, diagnostic, and recovery processes relating to certain ailments. Many approaches to monitoring a person's behavior in the home focus on instrumenting the individual in some way, such as using a smart watch or band, and trying to determine the types of activities in which the user is engaged, such as eating, sleeping, etc. This can be burdensome to the user as it requires vigilance to ensure the device is able to perform its task. We propose a method to unobtrusively monitor a persons movements within the home to generate an activity model through the use of a smart home sensor network. Using this model, we explore various methods to measure model differences that can be used to determine when an individual's activities deviate from an established routine. Our platform, the Automatic eXtensible Inferential Occupancy Monitor, or AXIOM, allows seamless data collection from multiple sensors as well as multivector predictive analysis using the generated activity model.