John Colias
John Colias
Senior Vice President at Decision Analyst
As a thought leader in advanced analytics, John focuses on predictive modeling, forecasting, and marketing research. As Senior Vice President with Decision Analyst, he helps his clients integrate data and modeling methods, draw valid conclusions, make better business decisions, and improve marketing effectiveness. John also teaches and conducts research as an Affiliate Assistant Professor of Business Analytics at the University of Dallas, where he is also Director of Master of Science in Business Analytics Program. His combination of academic and business interests helps analytics professionals to offer cutting-edge analytic solutions tempered by business realism. John holds a doctorate in economics from The University of Texas at Austin, with specializations in econometrics and mathematical modeling methods. He has been a frequent conference presenter of advanced modeling methods over the past 30 years.
Professional
Track: Machine Learning
Deliver Realistic Simulations by Incorporating Order Constraints Into ML Models
Machine learning models, especially deep learning models, deliver very accurate predictions of customer purchase decisions. However, when used for “what-if” simulations of price and feature changes, unrealistic outcomes may be produced. How would one improve the face validity and accuracy of simulations using machine learning models? This presentation will demonstrate a method to constrain parameters during the training process, improving the realism of simulation outcomes without sacrificing too much in the way of predictive accuracy. The listener will learn how to do “what-if” simulations with machine- and deep-learning models, incorporate constraints on layer parameters in neural networks, and deliver model insights through simulation. Examples and empirical results will be based on applying models with data from a publicly available dataset of store sales by customer.
