A Multi-Modal Approach to Patient Activity Monitoring

This paper presents our patient activity monitoring system (PAMS), which combines a wrist-worn device and a real-time location system (RTLS) with active tags. Four main data streams were used: location coordinates, 3D accelerometer, heart rate, and electrodermal activity (EDA) data. Using non-overlapping, normalized power windows, data are fed into random forest and KNN classifiers. These classifiers divide data into specific ambulation classes including time of inactivity, posture, and wheelchair monitoring data. The classification performance (F1 score) of our machine learning techniques ranges from 87.6% to 99.2% which is quite promising. These metrics provide clinicians with valuable insight into the activity levels of patients and patients' progress towards achieving their ambulation goals.