Abstract
This article presents a methodology for improving the performance of a trip generation/attraction forecasting model by means of the Box–Cox transformation. The application of this transformation to a set of random variables aims at obtaining a set of transformed variables distributed normally. The methodology comprises two steps. First, the sample is transformed employing the parameters of the transformation that maximises the likelihood as a multivariate normal distribution. In the second step, the forecast of the dependent variable, that is, the number of trips, is produced by an analytical approximation for the mean value of that variable conditioned on the rest of the variables. The methodology is applied to three different samples consisting of road trips and of a series of socio-economic variables. The analysis of the results shows that the model based on the optimal Box–Cox transformation has a better forecasting performance than that based on the logarithmic transformation.
Acknowledgements
This work is part of a research project financed by the Spanish Ministry of Public Works through R&D National Programmes (P 63/08 and P 62/08).