Enriching discrete choice models with computer vision for understanding choice behaviour in the presence of visual stimuli

Abstract

Discrete Choice Models (DCMs) are a key methodology in the transportation. However, current DCMs literally suffer from a blind spot: they cannot handle visual information. This blind spot hampers (1) a deeper understanding of human choice behaviour in the presence of visual stimuli and (2) using DCMs to deduce economic outputs for policies that involve changes to the visual environment. This study aims to bring visual information, in the form of images, to the realm of choice modelling. Specifically, it develops a series of discrete choice models –with computer vision parts embedded in different ways– to model the behaviour of decision makers when confronted with alternatives comprising both visual stimuli and conventional numeric attributes.