📊 Forecasting Health Service Demand

January 16, 2026 · View on GitHub

Binder
DOI License: MIT Python 3.11+ License: MIT

📊 Forecasting Health Service Demand

A practical course on forecasting methods for health service demand, part of the HPDM097: Making a Difference with Health Data module.

🚧 Note I maintain and update this repository once a year and update materials around December time.

📖 Overview

This repository contains hands-on materials for learning and applying forecasting techniques to healthcare data. Through six progressive computer labs, you'll master essential forecasting methods from naive benchmarks to advanced neural networks, all demonstrated with real-world health service applications.

What You'll Learn

  • Time series fundamentals and exploratory data analysis
  • Naive forecasting methods and benchmark models
  • Cross-validation strategies for time series
  • ARIMA model development and diagnostics
  • Facebook Prophet for daily/seasonal patterns
  • Neural network architectures for forecasting (Keras & PyTorch)
  • Best practices for model evaluation and selection

🚀 Quick Start

Prerequisites

  • Python 3.11 or higher
  • Basic understanding of Python, pandas, and NumPy
  • Familiarity with matplotlib for visualization
  • Basic statistics knowledge (recommended)

Installation

  1. Clone the repository

    git clone https://github.com/health-data-science-OR/forecasting.git
cd forecasting

  2. Create the conda environment

    conda env create -f binder/environment.yml
conda activate hds_forecast

  3. Launch Jupyter

    jupyter lab

📚 Course Structure

Get up to speed with essential date and time series handling:

  • Handling Dates: Working with temporal data in NumPy and pandas
  • Time Series Exploration: Visualizing trends, seasonality, and patterns with pandas and matplotlib

Lab 1: Foundations of Forecasting

Topics: Loading time series data, naive forecasting methods, baseline models

Materials:

  • Code-along: Loading time series into pandas
  • Code-along: Implementing naive benchmarks
  • Exercises: Naive forecasting practice

Key Skills: Data loading, simple forecasting models, establishing baselines

Lab 2: Model Validation

Topics: Time series cross-validation, train-test splits, forecast evaluation

Materials:

  • Code-along: Cross-validation techniques
  • Exercises: Implementing time series CV

Key Skills: Proper evaluation methods, avoiding data leakage, model comparison

Lab 3: ARIMA Models

Topics: Autoregressive Integrated Moving Average models, stationarity, model diagnostics

Materials:

  • Code-along: ARIMA introduction and implementation
  • Exercises: ARIMA modeling practice

Key Skills: Statistical forecasting, parameter selection, residual analysis

Lab 4: Facebook Prophet

Topics: Handling seasonality, holidays, trend changes in daily data

Materials:

  • Code-along: Prophet lecture and introduction
  • Exercises: Prophet application to health data; Revisiting ARIMA (Regression with ARIMA errors)

Key Skills: Automated forecasting, interpretable components, robust to missing data

Lab 5: Neural Networks Introduction

Topics: Feedforward networks, deep learning fundamentals, autoregressive architectures

Materials:

  • Code-along: Deep Learning 101
  • Exercises: Autoregressive networks in Keras (Part 1)
  • Optional: PyTorch implementation, OLS preprocessing

Key Skills: Neural network basics, model architecture design, training procedures

Lab 6: Advanced Neural Networks

Topics: Advanced architectures, hyperparameter tuning, ensemble methods

Materials:

  • Exercises: Autoregressive networks in Keras (Part 2)
  • Optional: PyTorch advanced implementation

Key Skills: Model optimization, regularization, production deployment

🛠️ Key Technologies

  • pandas & NumPy: Data manipulation and time series handling
  • matplotlib & seaborn: Visualization
  • statsmodels: ARIMA and statistical models
  • Prophet: Facebook's forecasting library
  • Keras/TensorFlow: Deep learning (primary framework)
  • PyTorch: Alternative deep learning framework (optional track)

📊 Example Applications

All examples use real-world health service scenarios:

  • Emergency department attendances, reattendances, and admissions forecasting
  • Ambulance dispatches
  • Forecasting monthly and daily level data

🤝 Contributing

Contributions are welcome! Please feel free to:

  • Report bugs or issues
  • Suggest improvements to exercises
  • Add additional examples or datasets
  • Improve documentation

📄 Citation

If you use these materials in your research or teaching, please cite:

@software{monks_2020,
  author = {Monks, Thomas},
  title = {Forecasting health service demand in python},
  year = 2020,
  publisher = {Zenodo},
  doi = {10.5281/zenodo.4332600},
  url = {https://doi.org/10.5281/zenodo.4332600}
}

📧 Support

  • Issues: Open an issue on GitHub for bug reports or questions
  • Module: Part of HPDM097 at University of Exeter
  • Author: Thomas Monks

📜 License

This project is licensed under the MIT License - see the LICENSE file for details. The course materials (text) are licensed under CC BY 4.0.

🌟 Acknowledgments

Special thanks to all contributors and students who have helped improve these materials.

Latest Release: v5.1.1 (January 16, 2026)