Graph Model based Activity Pattern Mining for Healthcare
Yongkoo Han, Ki-Sung Park, Young-Koo Lee · 2011
As the number of older people in the world population rapidly increases, personal and social costs related to healthcare are on rise. Recently a healthcare support system using long-term activity pattern monitoring is increasingly being studied as a new approach to support healthcare. The existing studies mine activity patterns based on statistics or sequential pattern model. Those models have shortcomings in that the statistics model cannot reflect all activity sequence and the sequential pattern model can mine only daily activity patterns. In this paper, we propose a notion of activity graph for activity pattern model and activity mining techniques using the activity graph. By creating activity graphs with different time windows. the proposed approach supports mining activity patterns with various periods such as daily, weekly, or monthly patterns. We adopt a multiple sequences alignment (MSA) for generating the activity graph with keeping each activity sequence information. In experiment, we show our proposed technique can generate more useful patterns for healthcare compared to existing activity pattern mining approaches.