Smartwatch based tumble recognition — A data mining model comparision study
Rainer Lutze, Klemens Waldhör · 2016
Tumbles are one of the key risk factors for elderly people. Studies have shown ([1]) that one third of all elderly above 65 tumble at least once a year with all its health related problems. In this study we present first a method for detecting tumbles using smartwatch sensor data and second evaluate various data mining models. The goal of this analysis is to find the most effective method for detecting tumbles in real time. We have compared nine different data mining models. Our investigations based on data collected from different ADLs and EDLs with a smartwatch show that three models have a comparable high recognition factor: neural nets, logistic regression and linear discriminant analysis while other methods perform less.