Isaac Sledge, James Keller, Timothy Havens, Gregory Alexander, Marge Skubic
When monitoring elders’ daily routines, it is desirable to detect aberrant activity trends, as they may foreshadow a need for medical attention. But, traditional, unsupervised pattern classification techniques are ill-suited for this task, because the data distributions formed by the captured patterns are temporal in nature. To overcome this algorithmic deficit, we craft a framework for analyzing and displaying additive trends in feature data extracted from passive sensors.
Submitted: Sep 11, 2008