Neural agent (Neugent) models of driver behavior for supporting ITS simulations

This paper presents an agent-based neuro-fuzzy approach for modeling drivers' compliance with travel advice under the influence of real-time traffic information. Fuzzy logic is combined with neural networks to capture the variability of drivers' appraisal of the different route attributes as well as the variability in their perceptions of the various attribute levels. The accuracy of the models, in terms of predicting the categories of drivers likely to comply with traffic advice, was found to exceed 90%. A comparative evaluation with discrete choice models showed higher accuracies ranging between (91 and 96) percent compared to (50-73) percent for the binary choice models