Machine Learning Applied to Walk-in Demand Prediction of a University Counseling Center

Abstract

University of Florida Counseling and Wellness Center (UF CWC) is one of the counseling centers that implemented a walk-in appointment policy for emergency needs. The attendance to UF CWC has increased in walk-in appointment traffic every year since data collection began in 2010, averaging a 7% increase in patient visits per year. However, demand for walk-in services is highly uncertain on an hourly, daily or weekly basis. Additionally, emergency needs of students should be met immediately before they become a catastrophic event. Thus, demand prediction becomes an important aspect to dynamically schedule counselors to deal with unexpected demand scenarios.  This project provides data visualization and utilizes machine learning techniques to predict future demand to assist with scheduling. We identified seasonal trends in historical visit data from the center, including peaks at the beginning of semesters and around finals. We then used the visit data to train a Gradient Boosting algorithm to predict demand. This model predicted demand with a mean of 4.2 patients per hour and mean square error of 1.75. Our results contribute to better demand prediction models for the UF CWC so that they may better support student needs with adequate staffing levels.