Dynamic Logistic Regression and Dynamic Model Averaging for Binary Classification

• Activity Number: 112
• Type: Topic Contributed
• Date/Time: Monday, August 2, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Tyler McCormick*+ and Adrian E. Raftery and David Madigan and Randall Burd
• Companies: Columbia University and University of Washington and Columbia University and Children's National Medical Center
• Keywords: Dynamic modeling ; Online classification

Abstract

We propose an online binary classication procedure for cases when parameters within a model change over time. We account for model uncertainty through Bayesian Model Averaging and allow model probabilities to also change with time. We apply a state-space model to the parameters of each model and a Markov chain model to the correct model, allowing our "correct" model to change over time. Our model accommodates different levels of change in the data-generating mechanism through a "forgetting" factor. A novel tuning algorithm which adjusts the level of forgetting in a completely online fashion using the posterior predictive distribution allows the model to accommodate various levels of change in the date-generating mechanism at different times. We apply our method to data from children with appendicitis who receive either a traditional (open) appendectomy or a laparoscopic procedure.