Applying an ensemble-based model to travel choice behavior in travel demand forecasting under uncertainties

ABSTRACT

The application of travel demand models to transportation planning has triggered great interests in issues that potentially improve the accuracy of model forecasts. These forecasts, however, are subject to various sources of input and model uncertainties. Focusing on travel choice behavior, this paper draws attention to the use of an ensemble-based model for addressing these uncertainties. A random multinomial logit (RMNL) model is developed by assembling a collection of multinomial logit (MNL) models. The bootstrapping procedure and the random feature selection are employed to capture the uncertainties in the model. A case study of investigating travel mode choice behaviors that illustrates situations necessitating the RMNL model is presented. Results suggest that the uncertainty related to predictions is reduced and the prediction accuracy is much improved. The RMNL model is computationally efficient and provides useful interpretations by estimating variable significance. Also, the RMNL model is able to deal with high-dimensional data.

KEYWORDS:

• Ensemble-based model
• travel mode choice
• bootstrapping
• random feature selection
• input uncertainty
• model uncertainty

Acknowledgments

This research is sponsored by the Research Grants Council of the Hong Kong Special Administrative Region (PolyU 152057/15E, PolyU 152095/17E), the Research Committee of The Hong Kong Polytechnic University (Project No. 4-ZZFY), and the National Natural Science Foundation of China (71801041, 71601052 and 71771049). Frank Witlox acknowledges that the research leading to these results received funding from the Estonian Research Council (PUT PRG306). The authors appreciate the helpful suggestions from Prof. William H.K. Lam at The Hong Kong Polytechnic University on this research.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

This work was supported by the Research Grants Council of the Hong Kong Special Administrative Region [PolyU 152057/15E, PolyU 152095/17E]; Research Committee of The Hong Kong Polytechnic University [4-ZZFY]; National Natural Science Foundation of China [71801041, 71601052, and 711771049]; Estonian Research Council [PUT PRG306].