Novel methodology development to address unsolved problems in prediction
When existing tools fall short, the SOUP Lab gets to work creating new ones. We develop novel statistical and machine learning methods to tackle some of the trickiest open problems in prediction modeling—like longitudinal/repeated measures data, missing data, model transportability, and meaningful uncertainty quantification. These are the kinds of issues that don’t always show up in textbook examples but are unavoidable in real-world applications. Our goal is to build methods that are not only theoretically sound but also practical enough to be used in applied settings—so they don’t just stay on the stove, they actually make it to the table. Whether it’s rethinking how predictions are made for individual patients or designing models that adapt to new environments, we focus on solutions that make prediction more robust, reliable, and ready for use.