Predicting Daily Patient Visits at Community Health Centers Using Machine Learning Algorithms to Optimize Medical Staff Scheduling

Abstract

Optimal health care at primary-level health facilities depends heavily on the readiness of human resources. At the Lempuing Community Health Center in Bengkulu, there is a significant operational imbalance due to unpredictable fluctuations in the number of daily patient visits. Medical staff scheduling has thus far been conducted conventionally without considering future trends in patient arrivals. This results in long lines on certain days, while also causing labor inefficiency during periods of low patient volume. This study employs the Cross-Industry Standard Process for Data Mining (CRISP-DM) methodology. The modeling approach utilizes the Random Forest algorithm. The computational process leverages historical data on daily patient visits over the past three months, specifically from October through December 2025. In addition to building a computational prediction model, this study also integrates the forecasting results with the Standard Workload (SBK) formula. Based on the results of the Random Forest model testing, the evaluation performance showed a Mean Absolute Error (MAE) of 7.27 patients per day and a Root Mean Squared Error (RMSE) of 10.25 patients. The coefficient of determination obtained was 37.10%, which reasonably indicates the level of complexity and the highly stochastic nature of historical health data, which is susceptible to random external factors. Feature importance analysis revealed that the H-7 and H-1 Visit variables are the most dominant parameters influencing spikes in patient numbers. The system is effectively capable of transforming digital prediction figures into recommendations for medical staff allocation, ranging from 1 to 3 per day.