Accurate forecasting of daily emergency department (ED) visit volumes is essential for staffing decisions, resource allocation, and patient flow management. While machine learning has demonstrated promise in forecasting ED demand for adult populations, its use in dedicated pediatric ED settings remains underexplored, particularly with respect to incorporating temporal and weather-related predictors. This study develops an XGBoost-based framework to forecast daily pediatric ED visit volumes one week in advance using ED visits data from a pediatric hospital between October 2021 and September 2025. The model incorporates lagged visit counts, rolling-window statistics, cyclical calendar encodings, and weather-related variables to capture recurring ED utilization patterns and environmental influences on ED visit volumes. The performance of the XGBoost model is benchmarked against three statistical time-series models: autoregressive (AR), autoregressive integrated moving average (ARIMA), and seasonal ARIMA with exogenous regressors (SARIMAX). On a one-year temporal holdout, XGBoost achieves a mean absolute error (MAE) of 13.54 and a mean absolute percentage error (MAPE) of 8.47%, reducing test MAE by approximately 21% and test MAPE by approximately 37% relative to the best statistical benchmark, SARIMAX (MAE = 17.18, MAPE = 13.50%). XGBoost also exhibits the smallest train–test gap among the evaluated models, indicating strong generalization across a full year of seasonal variation. Feature importance analysis identifies recent demand history, day-of-week cyclicality, and temperature variables as the most influential predictors. These findings demonstrate that XGBoost augmented with temporal and weather-related features meaningfully improves daily pediatric ED forecasting and provides a practical tool for short-term ED operational planning.