Modeling Epidemiological Time Series
Epidemiological data present naturally in a time series format. In many different contexts, these time series have structured biases due to the nature of the data generating processes. Identifying and correcting for these biases is crucial for accurate epidemiological modeling and forecasting. However, bias correction is a difficult task because biases vary by data source, there is limited access to historical data, and ground truth labels are almost always unavailable. In this thesis, we look at two classes of epidemiological time series: forecasts and real-time “indicators” of disease activity. For both classes, we describe the process of identifying biases and present different algorithms to correct them, depending on the context. We provide applications in modeling and forecasting influenza and COVID-19.
- Machine Learning
- Doctor of Philosophy (PhD)