Detection of abnormal sensor patterns in eldercare

In TigerPlace, an aging in place facility from Columbia, MO, we deployed 47 sensor networks together with an electronic health record (EHR) system to provide early illness recognition. In this paper, we describe a framework for predicting abnormal health patterns using non-wearable sensor sequence similarity. To compute the similarity between two sensor patterns we employ the Temporal Smith Waterman. A sensor pattern is classified as “abnormal” if it is much smaller than the mean of the distribution of “normal” patterns similarities. “Abnormal” days are defined by unusual sensor activity patterns that require a nurse's assessment of the resident. On a pilot data set of 1685 sensor days and 626 nursing records, we obtained a classification performance with an average precision of 0.70 and a recall of 0.30.