Bathroom activity monitoring for older adults via wearable device

Performance on activities of daily life (ADLs) is often used to evaluate older adults' capability of living independently. Current sensor technology has already accomplished classifying several types of ADLs automatically. However, thus far, limited studies were performed in bathroom activity monitoring. In this study, we proposed using one wearable accelerometer to monitor six types of bathroom activities: dressing, undressing, washing hands, washing face, brushing teeth and toilet using. One hidden-layer neural network model was applied for activity classification. The acceleration signal was separated into patterns, using 16s sliding window with 50% overlapping. The model was developed both with (a) 10-fold cross validation or (b) leave-one-person-out (LOPO) of the training set. The classification result was evaluated on precision, recall and F1-score. The F1-score was varied in these bathroom activities. In 10-fold cross validation, the highest average F1-score was of 91.19%, which was for toileting, and the lowest average F1-score was of 51.91%, which was for washing face. In LOPO, toileting still had the highest F1-score of 80.00%, but washing hands had the lowest F1-score of 9.62%. Additionally, we tested whether one participant's toileting event can be monitored in 3 consecutive days in the LOPO mode. The F1-score for classifying toileting events was 85.61%. In sum, the results show that it is possible to use one wearable accelerometer to monitor toileting activity. However, for the other bathroom activities, further study is needed and personal differences in these activities' performance should be considered.