Multi-model (Fusion) Systems

Abstract

As discussed in previous chapters, different models, taking different forms and constructed using different techniques, tend to display very similar performance when used to predict a wide range of consumer behaviors. However, sometimes (but not always) better forecasts can be achieved by constructing a suite of models and then combining the model scores together in some way. The terminology used to describe multi-model systems is diverse and many different terms are used to describe such systems. This includes: model ensembles, fusion systems, model combination and mixture of experts.