Pathways to improving traditional travel behavior models with travel-based multitasking and attitudinal data

Travel behavior;Multitasking;Activities while traveling;Mode choice;Millennials;Attitudes;Value of travel time;Transfer learning;Data fusion;Statistical matching;Machine learning;NHTS;Vehicle ownership

Malokin, Aliaksandr ; Mokhtarian, Patricia L. Civil and Environmental Engineering Circella, Giovanni Garrow, Laurie A. McCarthy, Patrick Mokhtarian, Patricia E. ; Mokhtarian, Patricia L.

University:Georgia Institute of Technology

Department:Civil and Environmental Engineering

关键词: Travel behavior;    Multitasking;    Activities while traveling;    Mode choice;    Millennials;    Attitudes;    Value of travel time;    Transfer learning;    Data fusion;    Statistical matching;    Machine learning;    NHTS;    Vehicle ownership;   

Others  :  https://smartech.gatech.edu/bitstream/1853/61259/1/MALOKIN-DISSERTATION-2019.pdf

美国|英语

来源: SMARTech Repository

【 摘 要 】

Transportation is undergoing extensive systemic changes: Information technologies are permeating through both transportation modes and people’s activity patterns, as vehicle automation and ride-hailing/sharing platforms are “catching by surprise” our tested and proven planning and forecasting tools, while lifestyle preferences and behaviors of millennials (the largest generational cohort in the U.S.) are accelerating the digitalization of travel and may be redrawing land use patterns. In these tumultuous and uncertain times, ever more pressure is put on travel behavior models to understand and predict travel patterns, and to provide a foundation for sensible decision making. Historically, regional transportation forecasting models used mostly socio-economic characteristics and relevant travel-related attributes to account for travel patterns. With the increased complexity and capacity for change of transportation systems, these factors could be insufficient for reliable policy and decision-making as the heterogeneity of travel preferences and experiences grows. Hence, the need for incorporating attitudinal data (an aggregate term for lifestyles, preferences, intentions, propensities, etc.), which underlies many travel-related decisions, into regional travel behavior models is especially strong now, and growing. Accordingly, the main goal of the present dissertation is to contribute to the improvement of regional travel behavior models by investigating the influence of understudied behavioral drivers and increasing the availability of attitudinal data. This goal can be decomposed into two distinctive parts, among other ways unified through the use of a single attitudinally-rich dataset: (1) studying the effects of travel-based multitasking on mode choice and the value of travel time (VOTT), and (2) developing an approach for porting attitudinal data from a small regional dataset to a large national sample. For the first part of the objective, the empirical analysis is based on a survey of Northern California commuters (N > 2,000) that measures travel multitasking attitudes and behaviors, together with other attitudes, mode perceptions, and standard socioeconomic traits. We estimate a revealed preference mode choice model, which accounts for the impact of multitasking attitudes and behavior on the utility of various alternatives. Results show that the propensity to engage in productive activities on the commute, operationalized as propensity to use a laptop/tablet, significantly influences utility and accounts for a small but non-trivial portion of the current mode shares. For example, the model estimates that commuter rail, transit, and car/vanpool shares would respectively be 0.11, 0.23, and 1.18 percentage points lower, and the drive-alone share 1.49 percentage points higher, if the option to use a laptop or tablet while commuting were not available. Additionally, the work investigates the differences between millennials and older adults in the sample. Compared to non-millennials, the mode choice of millennials is found to be less affected by socio-economic characteristics and more strongly influenced by the activities performed while traveling. For the second part of the objective, we transfer transportation-related attitudes from the same Northern California dataset to the 2009 National Household Travel Survey by augmenting both datasets with a large number of built-environment attributes and by applying machine-learning methods. Results indicate that the pro-transit, pro-active transportation, and pro-density attitudinal factor scores are predicted with the greatest precision; correlations of the predicted and observed scores are 0.564, 0.538, and 0.571, respectively. The performance of the transferred attitudes is measured by estimating linear regression models of vehicle ownership. The results show that in the source dataset the observed attitudes account for an 8.0% model lift (improvement in goodness of fit), while in the target dataset the predicted attitudes account for a 1.2–5.4% model lift. The present study presents the valuable combination of a novel empirical application together with a data augmentation methodology that could be transferred to a variety of contexts. To our knowledge, it is the first study based on a revealed preference model to quantify the contribution of travel multitasking attitudes and propensities to mode choice. Also, it is the first empirical study to occupy the intersection of three timely travel behavior topics: the impact of activities while traveling on mode choice, the estimation of willingness to pay and VOTT, and the analysis of the travel behavior of millennials. Finally, this work acknowledges that with many transportation planning decisions requiring large-scale comprehensive datasets to be fed into travel behavior models, it would be difficult to achieve the introduction of a novel class of variables (e.g., activities while traveling) into the existing modeling pipelines. The proposed transfer learning framework targets this data unavailability and offers a way to synthesize promising variables into a practice-ready context.