Improved Trip Distribution Models Using the Empirical Bayes Method

The destination choice model considers each individual as a fundamental analytical unit, thereby providing increased flexibility in accommodating a wide range of factors that influence destination choice behaviors compared to aggregate trip distribution models. However, in practical applications, practitioners have found that there are still influencing factors that cannot be represented in the model, resulting in a huge deviation between the predicted trip distribution and the actual trip distribution. To enhance the predictive accuracy of the destination choice model, this paper proposes a new computational method based on a combination of the traditional destination choice model and the empirical Bayes (EB) method. It can integrate additional information from trip frequencies in the survey data with the information from various influencing factors in the traditional destination choice model. The proposed EB model is developed for commuting trips based on 2019 Shanghai Household Travel Survey data, and the model is validated using the survey data and mobile signaling data. For comparisons, both the traditional destination choice model and the proposed EB method are applied to distribute trips. The analytical results indicate that the EB method outperforms the traditional model in all the goodness-of-fit indicators, and the prediction errors are reduced in all the validation procedures. It is therefore expected that the proposed EB method based on Bayesian statistics may improve the predictive capability of the destination choice model for trip distribution.