Method of behavior modeling for detection of anomaly behavior using hidden Markov model

Abstract

Dementia is disorder of memory and judgment caused by dying brain cells. Most of dementia symptoms can be only detected by housemates when they notice some changes in behaviors of elderly people. Therefore, it is difficult to detect the early symptoms of dementia in elderly people living alone. We focused on wandering which is typical symptom of dementia. We proposed the system to detect wandering symptom based on sensors data using Hidden Markov Model (HMM). We installed sensors to acquire elderly people behavior. Then, we created normal behavior model based on these sensors data using HMM. After that, we compare between this pattern model and detected behavioral pattern. When the detected behavioral pattern did not much with this pattern model, the system will classify the behavior as a wandering symptom. In this paper, we proposed the method which is a creation of normal behavior model. We verified whether our proposal method can estimate behavioral tendency of healthy elderly subject's living alone using two months' sensor data. The result suggested that this method can create a behavior model with considering about subject's habits.