Analysis of Feature Importances for Automatic Generation of Care Records

Abstract

The aging society is becoming a more and more serious problem in developed countries. In Japan, the aging society is also becoming a serious problem same as other developed countries. As a result, the shortage of caregivers is expected to be a severe problem. For this reason, Information and Communication Technology (ICT) and other technologies are expected to be used to reduce the workload of caregivers. In our laboratory, we have developed a nursing care record application which used for recording the nursing care activities and nursing care records. In this paper, to improve the estimation accuracy of nursing care records, we analyze the nursing care record data and machine learning model using feature importance and data visualization. We proposed 10 new features e.g., previous amount of breakfast, previous amount of lunch, previous amount of dinner, snacking, level of requiring long-term care, sleep time on the previous day, amount of exercise on the day, physical condition on the day, weekday, and weather. Afterwards, we analyze the trends of the care record data and estimation results based on these new features. The evaluation of the average accuracy was 77.3% and the average F1 score of 49.8%. CCS CONCEPTS• Human-centered computing → Ubiquitous and mobile computing.