Predicting Outpatient Appointment Demand Using Machine Learning and Traditional Methods.

Area of Science:

• Healthcare Operations
• Data Science
• Predictive Analytics

Background:

• Traditional forecasting methods are widely used for clinical demand.
• Machine learning offers advanced capabilities but presents challenges in model selection and optimization for accurate forecasting.

Purpose of the Study:

• To compare the predictive accuracy of machine learning and traditional methods for outpatient appointment demand.
• To identify the optimal forecasting model for predicting patient appointment requests.

Main Methods:

• A retrospective analysis of appointment requests from two outpatient locations (A and B) was conducted.
• Twenty traditional, hybrid, and machine learning models were evaluated using Forecast Standard Error (FSE) as the primary metric.
• Data characteristics, including cyclical and trending patterns, were examined for each location.

Main Results:

• The feature-engineered XGBoost model, a machine learning approach, achieved the lowest out-of-sample FSE at both Location A and Location B.
• Location A's data exhibited a cyclical, non-trending pattern, while Location B's data showed a cyclical and trending pattern.
• The performance of forecasting models varied based on underlying data characteristics.

Conclusions:

• Understanding data patterns is essential for selecting appropriate forecasting methods.
• Machine learning models, particularly when enhanced with feature engineering or hybrid approaches, can significantly improve the accuracy of clinical demand forecasting.
• A comprehensive evaluation of diverse forecasting models is recommended to achieve optimal predictive results.