2016 Volume 3 Issue 1 Pages 21-30
The need for home monitoring systems for the elderly is growing because the number of the elderly living alone is on the rise. We propose an anomaly detection method in a home monitoring system for the elderly that takes a time-series of the subject’s activity level data captured by simplistic motion sensors as its input. First, the activity level data were classified into four types based on the trend and volatility of the data of 144 elderly people. Second, we then applied financial models that are suitable for each type of data in order to detect anomalies. For example, an underlying asset pricing model was applied to the data that exhibited trend-presence and constant-volatility. Here an anomaly was defined to be a datum that lies outside the 95% confidence interval for the given model. Finally, the feasibility of this method was evaluated by applying it to the data from a nation project on the activity of elderly people.
