Predicting Purchases, Rare Diseases, and More: Using Ordinal Regression to Estimate Rare Event Probabilities
Estimating probabilities of rare events like purchases from click ads or contracting rare diseases is difficult because of class imbalance. In this talk, I will show how ordinal regression can make these estimation problems easier. I will explain the methods step-by-step, show you real data examples, and share code implementing the methods.
Whether predicting online purchases from an ad click, diagnosing rare diseases, or serving tailored recommendations, the ability to accurately estimate the probability of rare events is key for exceptional data-driven decision making. The challenge? Class imbalance, a widespread problem where rare positive examples become the needle in the data haystack. In this talk, I will show how ordinal regression can be used to make this problem easier, leading to more effective ad spending, better disease diagnosis, stronger recommendations, and more. I will start by walking through simpler models and build up to a cutting-edge method I developed that was accepted at this year's International Conference on Machine Learning (ICML 2023). Along the way I will share code implementing these methods and show you how it works in real data examples. Come to see how you can learn more from your organization's data and find better solutions for important problems.