Forecasting patient visits in primary healthcare using arima, holt's exponential smoothing, prophet, and weighted ensemble models

Abstract

The variability of patient visits in primary healthcare facilities requires accurate and interpretable forecasting methods to support service planning, resource allocation, and staff scheduling. This study aims to evaluate the forecasting performance of ARIMA, Holt’s exponential smoothing, Prophet, and weighted ensemble models in predicting monthly patient visits at XYZ Primary Clinic in Bogor City. A quantitative time series forecasting approach was applied using monthly patient visit data from January 2020 to December 2023, consisting of 48 observations. The data were divided into 38 training observations and 10 testing observations using an approximately 80:20 split. Model performance was evaluated using Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE), and Mean Absolute Error (MAE). The results show that Holt’s exponential smoothing provided the best individual forecasting performance, with the lowest MAPE of 9.39%, RMSE of 169.43, and MAE of 141.60. Among the ensemble configurations, the combination of Holt’s and Prophet produced the best ensemble performance, with a MAPE of 9.56%, RMSE of 171.01, and MAE of 146.11. However, the weighted ensemble models did not outperform the best individual model. These findings indicate that the patient visit data exhibit a relatively stable trend pattern, making Holt’s exponential smoothing more suitable than more complex approaches for this dataset. This study highlights the importance of selecting forecasting models based on data characteristics rather than model complexity alone. The findings provide practical implications for improving patient visit forecasting, resource planning, and service management in primary healthcare settings.