In mode-choice modelling, the realistic understanding of individual behaviour requires an explicit integration of various contextual factors, as well as variables describing individual perception, attitudes, and situational constraints. However, such external factors are often missing (or only implicit) in conventional econometric modelling. On the other hand, phenomenological approaches based on in-depth comprehension of behaviour are seldom operational. The new approach presented in this chapter, SATCHMO, aims at going beyond this contradiction. By using a constraint-based logic, this disaggregated knowledge-based system attempts to reproduce and restrain individual choice universes automatically, detecting captivities and, if possible, determining actual choices. It formalises the situational and objective constraints system (socioeconomic, geographic, temporal, vehicle availability) explicitly, that affect usual mode choices made by commuters, by taking into account interrelations within households. Two kinds of rules are applied to each fact base representing a household: first, certain rules aim at reducing the universe of modes of each commuter; and, second, uncertain rules aim at estimating choice probabilities. SATCHMO's framework is described and one of its possible applications is presented. Using data from any classical household survey, SATCHMO allows us to eliminate, with very little error, some of the possible choices, ending in a formalised reliable system for the restriction of choice sets, i.e. for the manifestation of captivity. Therefore, such a tool can be used in two different ways, either as an independent simulator, or as a complement to a logit model.
Abstract