Machine learning for product choice prediction

Abstract The goal of this paper is to provide a point empirical evidence as how machine-learning techniques stack-up in their ability predict consumer choices relative traditional statistical techniques. We compare (naïve) multinomial logit six alternatives: learning logit, random forests, neural networks, gradient boosting, support vector machines and an ensemble algorithm. comparison done by applying these methods beer category stock keeping unit (SKU) level panel data. Results show that tend perform better, but not always. Ensemble performs best while maintaining overall high-performance across all SKU classes, independently sample size. This result builds on existing about the benefits combining multiple prediction over relying single performing model, conventional wisdom would intuitively make us believe. In general, better performance machine at predicting product choice should come surprise. At core, are designed augment dimensionality models and/or scan through orders magnitude greater model alternatives, narrower focus approaches.