Self-Validated Ensemble Modeling: Improving Prediction from Designed Experiments | Fireside Chat

Topic Overview This session explores Self-Validated Ensemble Modeling (SVEM), a novel approach for improving prediction from designed experiments. Traditional methods can struggle with small experimental datasets that cannot easily be divided into separate training and validation sets. SVEM addresses this challenge by using a unique weighting approach that allows observations to contribute to both model training and validation. The presentation examines SVEM with several model-building methods, including stepwise regression, LASSO, and the Dantzig selector. Through simulations and case studies, the session demonstrates how SVEM can produce models with better predictive performance than traditional one-shot model selection approaches. Thought Leader: Maria Weese is Professor of Information Systems & Analytics, Faculty Fellow, and MSBA Faculty Director in the Farmer School of Business at Miami University. Her research focuses on the design and analysis of experiments, statistical monitoring, and applications of analytics and machine learning. Prior to entering academia, she worked as a Process Improvement Engineer at Celanese Acetate. Dr. Weese earned her PhD and MS in Statistics from the University of Tennessee and her bachelor’s degree in Chemical Engineering from Virginia Tech.