Nursing Activity Recognition and Future Prediction in Hospitals

Sozo Inoue,
International Symposium on Applied Engineering and Sciences (SAES2016)
(Not Available)
(Not Available)
Recently, researchers have explored the possibility of human activity recognition with mobile sensors.
In this paper, we firstly introduce our work where we proposed a method for recognizing whole day activities using prior knowledge of the information of a sequence of activity segments which are obtained from a whole day training dataset, such as the daily timestamps, duration, and imbalances among activity classes. To evaluate the method, we collected actual activities from nurses wearing accelerometers in a hospital for approximately 2 weeks and combined them with training labels, which resulted in 25 activity classes with 5,743 labels from 22 nurses. As a result, the proposed method outperformed the naive method without using prior knowledge by 25.81% at maximum through the balanced classification rate.
Secondly, we extend the research of understanding nursing activities to analyze near futures integrated with medical records. We carried out an experiment to collect nurse activity and location data in cooperation with one floor of a hospital, which constitutes the orthopedic surgery department, for 40 days, 24 hours per day. In addition, we collected medical data such as DPC data (for calculating medical payments), patient status data (such as nursing needs), and number of hospitalization days.
Utilizing the collected data, we predicted the nursing load of the next day from patient status of a given day. As a result, the accuracy of predicting whether the next day’s nursing time for a patient will be longer based on the previous day’s patient status was 73.7%.

Data Files

No records to display