One of the main benefits of a wrist-worn computer is its ability to collect avariety of physiological data in a minimally intrusive manner. Among thesedata, electrodermal activity (EDA) is readily collected and provides a windowinto a person's emotional and sympathetic responses. EDA data collected using awearable wristband are easily influenced by motion artifacts (MAs) that maysignificantly distort the data and degrade the quality of analyses performed onthe data if not identified and removed. Prior work has demonstrated that MAscan be successfully detected using supervised machine learning algorithms on asmall data set collected in a lab setting. In this paper, we demonstrate thatunsupervised learning algorithms perform competitively with supervisedalgorithms for detecting MAs on EDA data collected in both a lab-based settingand a real-world setting comprising about 23 hours of data. We also find,somewhat surprisingly, that incorporating accelerometer data as well as EDAimproves detection accuracy only slightly for supervised algorithms andsignificantly degrades the accuracy of unsupervised algorithms.
展开▼