Travel Behavior Classification: An Approach with Social Network and Deep Learning
Yu Cui, University at BuffaloShow Abstract
Qing He, University at Buffalo
Alireza Khani, University of Minnesota, Twin Cities
Uncovering human travel behavior is crucial for not only travel demand analysis but also ridesharing opportunities. To group similar travelers, this paper develops a deep learning based approach to classify travelers’ behaviors given their trip characteristics, including time of day and day of week for trips, travel modes, previous trip purposes, personal demographics, and nearby place categories of trip ends. This study first examines the dataset of California Household Travel Survey (CHTS) between the year of 2012 and 2013. After preprocessing and exploring the raw data, we construct an activity matrix for each participant. The Jaccard similarity coefficient is employed to calculate matrix similarities between each pair of individuals. Moreover, given matrix similarity measures, we construct a community social network for all participants. We further implement a community detection algorithm to cluster travelers with similar travel behavior into the same groups. There are five clusters detected: non-working people with more shopping activities, non-working people with more recreation activities, normal commute working people, shorter working duration people, later working time people, and individuals needing to attend school. We further build an image of activity map from each participant’s activity matrix. Finally, a deep learning approach with convolutional neural network is employed to classify travelers into corresponding groups according to their activity maps. The accuracy of classification reaches up to 97%. The proposed approach offers a new perspective for travel behavior analysis and traveler classification.
An Explorative Analysis of Group Travel Behavior Patterns in the Public Transit Context
Yongping Zhang, University College LondonShow Abstract
Karel Martens, Technion - Israel Institute of Technology
This paper explores group travel behaviour (GTB), which is defined as two or more persons intentionally traveling together during (part of) a trip. Most travel behaviour studies are based on the assumption of isolated individuals and ignore interpersonal relationships between travellers. Here we propose a smart card data-based method to identify GTB, drawing on insights from the theory of proxemics, which is widely used to describe the role of interpersonal distance between individuals in Social Psychology. This method uses interpersonal time distance, namely the time interval between persons’ enterance/exit time, to distinguish GTB from persons traveling on their own. We apply the method using smart card data generated by the Shanghai metro system, and find that the value of interpersonal time distance strongly shapes the share of travel that is classified as GTB. However, when the interpersonal time distance is small enough (e.g., smaller than 10 seconds in the case of Shanghai), it is possible to reveal overall GTB patterns at a metropolitan level and at a fine temporal scale with a reasonable level of certainty. Results from our Shanghai study also show that the GTB pattern is distinctly different from the pattern of individual travel in terms of both time and space. For example, Group trips tend to occur in weekends, in the afternoons and evenings. Stations close to the central city and serving more leisure-related destinations show higher GTB shares. While acknowledging the inherent uncertainties related to the approach, we argue that the proposed smart card data based method is a straightforward and workable solution to reveal GTB patterns at a large spatial scale and at a fine temporal resolution and a promising approach that can be applied in more situations.
Studying the Relationship Among Activity Participation, Social Networks, Expenditures, and Travel Behavior on Leisure Activities
Maximiliano Lizana, Universidad de la FronteraShow Abstract
Juan Carrasco, Universidad de Concepcion
Alejandro Tudela, Universidad de Concepcion
In the context of the increasing interest on non-mandatory activities – such as those related to recreation and socializing – this work focuses on studying the relationships between participation in activities, social networks and expenditures in daily travel patterns associated to leisure activities, as a way to understand the people’s strategies to perform activities in daily life. The methodology included the use of a daily activities log applied during seven days, along with a socio-demographic and social network characterization. Using Structural Equations Models, the study provides empirical evidence of the effect of individual social networks on people’s travel patterns. The results suggest a positive relationship between the interaction of people with their social networks and the level of expenditure with respect with people’s time and spatial activity patterns. The analysis contributes towards our better understanding and modelling of people’s behavior in time and space and the role of social networks and expenditures associated to different daily activities.
Effect of Social Influence on Consumer Choice Behavior Using a Sequential Stated Choice Experiment: A Study of City Trip Itinerary Choice
Xiaofeng Pan, Eindhoven University of TechnologyShow Abstract
Soora Rasouli, Eindhoven University of Technology
Harry Timmermans, Eindhoven University of Technology
This paper introduces a model that captures the effect of social influence on individual choice behavior. The suggested model shares with previous models the idea to add a term to the deterministic utility function of the choice alternative, chosen by a social network member, to measure an individuals’ sensitivity to social influence. To account for individual differences in sensitivity to social influence, this term is assumed a function of the individual’s socio-demographic profile and the nature of the relationship between the individual and the social network member, differentiating between friends, relatives and neighbors. The model generalizes prior models on social influence from binary to multi-alternative choices. To assess the validity of the model, the choice of city trip itinerary was chosen as an example. A sequential stated adaptation experiment, in which individuals first choose the itinerary they personally like best from a series of experimentally varied choice sets and then choose again from the same choice sets after being informed about the choice of a social network member, constitutes the basis of the model estimation. Results show that the model reproduces observed data well. Social influence has a modest but significant positive effect on individuals’ choice behavior. In addition, the strength of social influence is shown to have a significant positive relationship with income (≥ 3126 Euro/monthly) and particularly if the social network member is a friend.
Intrahousehold Interaction on Daily Maintenance Activity Stop Generation: A Copula-Based Approach
You-Lian Chu, ParsonsShow Abstract
This paper presents a stop frequency model used to predict the daily number of maintenance activity stops made by individual household heads during a typical weekday. The model simultaneously considers three activity settings: male head undertakes activity independently, female head undertakes activity independently, and both heads undertake activity jointly. To account for interactions between household heads and the ordered classification scale of the dependent variable (in this case, number of stops - zero stop, one stop, and two or more stops), a modified multivariate ordered probit model was used to model activity generation within households; that is, the proposed model maintains the probit assumption for the marginal distributions while introducing non-normal dependence among the error terms using copula functions.
Using the New York Metropolitan Transportation Council’s 2010/11 Regional Household Travel Survey data, the copula-based multivariate ordered probit model was employed to examine stop-making behavior for individual household heads residing in New York City and its adjacent counties in Mid-Hudson Valley and New Jersey. The empirical results provide useful insights into the observed effects of socio-demographics, land use density, transportation service, and work schedule together with potential unobserved common effects on individual stop-making propensity in a highly-urbanized environment. They also demonstrate that using a copula-based approach to evaluate the inter-relatedness of spousal stop-making decisions at the household level is worthwhile
Modeling the Loss and Retention of Contacts in Social Networks: The Role of Tie Strength and Dyad-Level Heterogeneity
Chiara Calastri, University of LeedsShow Abstract
Stephane Hess, University of Leeds
Andrew Daly, University of Leeds
Juan Carrasco, Universidad de Concepcion
Charisma Choudhury, University of Leeds
The social network of an individual has substantial impact on the activity, travel and other choices, both directly (e.g. by means of joint activities and travel) and indirectly (e.g. by means of its influence on shaping up attitudes and perceptions which in turn affects decisions. In order to capture the impact of social network in choice models, there is a strong need for modelling how the social network forms, evolves over time and what factors contribute to the loss/retention of the social contacts. There is a distinct research gap on the literature, with existing work failing to capture the full extent of ego-level and ego-alter level heterogeneity in the loss/retention processes. We propose the use a hybrid model framework which is based on the notion of a latent strength of relationship. We demonstrate the potential of the approach using data from Chile, showing the presence of extensive variations both at the ego and ego-alter level, only some of which can be linked to observed characteristics.
Providing Personalized Feedback to Investigate the Role of Social Influence on Travel Behavior
David Palma, University of LeedsShow Abstract
Romain Crastes dit Sourd, University of Leeds
Chiara Calastri, University of Leeds
Stephane Hess, University of Leeds
Vikki O'Neill, Queens University
Promoting active travel is beneficial from both a public health perspectiveand a transport management one. Providing feedback to travellers about their travel behaviour is a popular method to encourage active travel and reduce the use of car. Past studies that have measured the effect of feedback on mode choice and distance travelled by different modes have done so using simplifying assumptions, i.e. measuring its effect independently of the nature and directionality of the feedback provided, and often only providing one type of feedback, for example miles driven by car or calories burnt.
Using a simple linear panel approach, we measure the effect of two types of feedback on a group of travellers during a two-week period. In particular, in addition to a control group, one treatment group receives feedback only about their own travel activity, while people in another group are also provided with information about how they compare with others.
We measure the effect of being in the different treatment groups on several outcomes, such as calories burnt, CO2 emitted and distance travelled by active and motorised modes. We find significant differences, with bigger reductions in car use among those who receive information about others. We also find the feedback effects to be asymmetrical with respect to a reference, in such a way that only those who pollute more than others reduce their emissions. In summary, our findings offer additional insight into how people react to different types of feedback, providing a valuable tool for sustainable transport policy.
Endogenous Effects of Social Network on Destination Choice in Neighborhoods in Hanoi, Vietnam
Hong Nguyen, Hiroshima UniversityShow Abstract
Makoto Chikaraishi, Hiroshima University
Akimasa Fujiwara, Hiroshima University
Junyi Zhang, Hiroshima University
Effects of social network on travel behavior have been widely examined in literature; however, there is little study focusing on the impacts of social network in a neighborhood on destination choice, though destination choice can be significantly influenced by social interaction effects among neighbors. Also, there is a little empirical study on identifying social interaction effects in the context of Asian developing countries. To fill in the above research gaps, this study examines the endogenous effects of social network in a neighborhood on destination choice in Hanoi Metropolitan Area of Vietnam. Specifically, a logit-based destination choice model is developed, where the impacts of social network is represented by the average destination choice probability of his or her acquaintances. Two methodological challenges are addressed in this paper: (1) observing the whole social network in a neighborhood is not feasible, and thus it is simulated based on the partial information of social network, and (2) a structural estimation method under the control of unobserved neighborhood-level characteristics is adopted in order to reduce biases in the estimated social interaction effects. Using data collected at three new neighborhoods in Hanoi in 2015, we confirm that the ignorance of unobserved group effects would cause biases in the parameter of endogenous social interaction effects, while the modest social interaction effects on destination choice behavior would exist.