Applicability of Genetic Algorithms for Estimation of Mode Choice Models
Conference Paper
Overview
Research
Identity
Additional Document Info
View All
Overview
abstract
This research explored the use of Genetic Algorithms (GA) to develop a mode choice model for a dataset that included an extremely complex set of mode choices. The data used were stated preference survey data collected from travelers on Katy Freeway in Houston, Texas. Respondents had nine travel modes with different times of day (peak versus off-peak), occupancy and toll options to choose from. This paper attempts to use GA for estimation of Multinomial Logit (MNL), Nested Logit (NL) and Random Parameter Logit (RPL) model coefficients. Log likelihood function for each of the model was considered as the fitness function for the genetic algorithm estimation. Different optimization options such as population size, selection function, mutation, etc. available with the genetic algorithm were examined to understand their impact on coefficient estimation and develop the best model. The log-likelihood values for the models estimated using GA were compared to those obtained for models estimated using traditional analytical methods. Traditional estimation techniques may lead to suboptimal solutions if used for estimating models with complex log likelihood function such as the nested logit or random parameter logit model. GAs have been found useful for solving the problems with multiple optima and complex structure. This study finds that it is possible to estimate complex mode choice models using GA. Copyright 2008 by the International Institute of Informatics and Systemics.
