Change Point Detection Using Multivariate Exponentially Weighted Moving Average (MEWMA) for Optimal Parameter in Online Activity Monitoring

In recent years, wearable sensors are integrating frequently and rapidly into our daily life day by day. Such smart sensors have attracted a lot of interest due to their small sizes and reasonable computational power. For example, body worn sensors are widely used to monitor daily life activities and identify meaningful events. Hence, the capability to detect, adapt and respond to change performs a key role in various domains. A change in activities is signaled by a change in the data distribution within a time window. This change marks the start of a transition from an ongoing activity to a new one. In this paper, we evaluate the proposed algorithm’s scalability on identifying multiple changes in different user activities from real sensor data collected from various subjects. The Genetic algorithm (GA) is used to identify the optimal parameter set for Multivariate Exponentially Weighted Moving Average (MEWMA) approach to detect change points in sensor data. Results have been evaluated using a real dataset of 8 different activities for five different users with a high accuracy from 99.2 % to 99.95 % and G-means from 67.26 % to 83.20 %.