Estimation of Office Worker’s Emotions Using Wearable Biometric Sensor

抄録

We developed a model that estimates the emotions of office workers using information obtained by wearable biometric sensors. In 11 healthy office workers, pulse rate, pulse rate variability (PRV), skin temperature, body motion, and conversation time were continuously monitored between 08:30 and 21:30 on every workday using a bracelet-shape wearable sensor. During the monitoring, subjects recorded four strongly conscious emotions (happy, relaxed, sad, and angry) every 30 min. Based on the Russel’s Circumplex emotion model, four emotion types were developed into coordinates consisting of arousal and valence axes. The linear models were developed for estimating the value of each axis using the biometric information. From a total of 911 days of records in the 11 subjects, a total of 9,737 hours of data were obtained. By stepwise regression analyses, the coefficient of variation of high-frequency PRV amplitude and skin temperature were extracted as the best variable combination explaining the valence axis, and the frequency variability of PRV respiratory peak and conversation time were extracted for the arousal axis. The models discriminated between high valence (happy and relaxed) and low valence (angry and sad) states with an AUC of receiver-operating characteristic curve of 0.64 (P = 0.0001) and discriminated between high arousal (happy and angry) and low arousal (relaxed and sad) states with an AUC of 0.61 (P = 0.001). Our findings suggest that information related to two coordinates comprising the Russel’s Circumplex model could be obtained by wearable biometric sensors, helping the estimation of emotion type of office workers.