Clustering action data based on amount of exercise for use-model based health care support

This paper proposes construction of user-model from mining the action log and amount of activity using Neural Network and EM algorithm. User model is useful tools for providing information or services which are suit each user. The authors focused on user-model base on action log and amount of activity. The propose a method scores the action data based on life rhythm or amount of exercise, and weighs the scored data using the weights which is calculated by neural network. The method, finally, clusters the scored data and exercise intensity (METs) using EM algorithm. This paper shows 2 applications. First application uses collected action log and amount of activity from action records which is recorded by participants. The action records include user's action, amount of exercise (METs), location, and transportation. These data are scored base on life rhythm. Second application uses 3D accelerometer with motion recognition system, and pedometer which can be measure METs every 1 min. These data are scored exercise intensity or duration of the exercise. As shown in results of 2 applications, the system finds 3 or more clusters from these data. Each cluster reflects users' exercise data.