synthetic-ae-demand-xgboost

synthetic-ae-demand-xgboost

This model is trained entirely on synthetic data. It is a demo/portfolio project showing an end-to-end XGBoost forecasting pipeline, styled after an A&E (emergency department) daily-demand forecasting problem. No real patient or NHS data was used anywhere in this project.

Model description

XGBoost regressor predicting a synthetic daily "attendance" count from calendar features and rolling averages.

Features: day of week, day of year, weekend flag, synthetic-holiday flag, 7-day and 28-day rolling averages.

Training data

Synthetic daily time series generated locally with weekly seasonality (higher on weekends), annual seasonality (winter peak), a mild upward trend, occasional synthetic "holiday" spikes, and Gaussian noise. Generation code is included in this repo (generate_synthetic_ae_data) so the data is fully reproducible and inspectable.

Evaluation results

(on a chronological 80/20 train/test split of the synthetic data)

• MAE: 6.271
• RMSE: 8.163
• R²: 0.83
• Train / test size: 1200 / 300

Intended use & limitations

Demo/portfolio use only. Because the training data is synthetic, this model has no predictive value for real-world emergency department demand and should not be used for anything operational or clinical.

How to use

import xgboost as xgb

model = xgb.XGBRegressor()
model.load_model("model.json")
preds = model.predict(X)  # X must have the same feature columns as training