1002010 - Forecasting in Public Health

Course Description

This course introduces students to modern time series analysis. Students will first develop a solid understanding of the foundations of time series analysis, including data creation processes, forecasting strategies (recursive, direct, and multi-output), forecasting types (point and probabilistic), evaluation and scoring techniques, and key families of forecasting models, among others. The course then extends these concepts to real-world applications in epidemic forecasting. The primary focus is on forecasting using statistical methods, deep learning approaches, and time series foundation models (TSFMs). Additionally, we will briefly highlight the role of compartmental models and hybrid methods and compare them with data-driven methodologies.

Learning Outcomes

By the end of the course, you will be able to:  

1. Handle time series data.  

2. Apply classical statistical models for time series forecasting (e.g., ARIMA, ARIMAX, Prophet).   

3. Implement and evaluate machine learning and deep learning models for sequential data.   

4. Compare different modeling approaches using appropriate evaluation metrics (e.g., MAE, RMSE, MAPE, PIC, WIS).   

5. Critically assess the strengths and limitations of different forecasting approaches, including modern foundation models (e.g., zero-shot forecasting methods).   

6. Communicate analytical results clearly in a structured technical report. 

Assessment

Microcredential badge earned upon successful completion