Submission from Team Alpha, Applied AI Lab at UNC Wilmington, for 4th Nurse Care Activity Recognition Challenge
As people live longer, health care services for the elderly have increased in demand. Applying human activity recognition (HAR) to the field of nursing can reduce the workload of nurses and other healthcare workers by keeping track of which patients have been treated and at what time. For the 4th Nurse Care Activity Recognition Challenge, hourly predictions of nurse behavior were generated by training models with temporal data, such as the hour or date of the activity. The data was collected in May and June 2018 on a smartphone, which remained in the nurse’s pocket as they performed their daily activities at a healthcare facility. This paper provides and analyzes the results from our group, Team Alpha. Two models were tested, Random Forest Classifier, and K-Nearest Neighbors. Of the two models, the Random Forest Classifier achieved better results with an average F1-score of 59.2% in classifying 28 activities performed across five nurses.
Python, Pandas, Matplotlib, Jupyter Notebook, Scikit-Learn