Elevating Inpatient Admissions Forecasting Through Sequential Feature Inclusion

The dynamic pattern of inpatient admissions at any hospital is a critical area of concern as it deals with managing with limited resources. Accurate forecasting is crucial in such situations. This study attempts to predict monthly inpatient hospital admissions using the data of a district hospital from January 2021 to June 2023 using three machine learning models: random forest (RF) regressor, support vector regression (SVR), and extreme gradient boosting (XGBoost); and with two conventional forecasting models: seasonal autoregressive integrated moving-average (SARIMAX) and linear regression (LR). This study aims to discern the most precise model for predicting monthly inpatient admissions. Employing Shapley additive exPlanations (SHAP) scores, the study seeks to identify the most influencing features impacting the prediction of inpatient admissions. The findings reveal that the XGBoost model emerges as the most accurate model, followed by the RF Regressor and SVR. SARIMAX ranks as the least accurate among the models considered. Notably, the rolling average, with a window size of 7, exerts the most significant influence on inpatient admissions, followed by the day of the week, lag factors, and temperature.