Prediction of BPSD Using Environmental and Vital Sensor Data

Behavioral and psychological symptoms of dementia (BPSD) that develop in patients in nursing homes or living at home impose a heavy burden on caregivers and family members. In this study, we collect data from environmental and vital sensors at multiple nursing homes. Based on the data, we propose a BPSD prediction method using gradient boosting trees, a machine learning technique. An evaluation experiment, using data gathered over approximately 10 months, achieved an AUC of 0.7217 for the ROC curve and an Average Precision of 0.3961 for the PR curve. The results show the potential of machine learning methods using environmental and vital sensors to predict BPSD.