Travel Behavior Analysis and Mode Choice Prediction for Commuting to Campus - Performance Comparison of Discrete Choice and Machine Learning Models
Date
2021-09-21
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Extensive studies exist on various aspects of travel behaviour of the general population; however, few works explore the commuting behaviour of university commuters. This thesis investigates university commuters' commuting habits/attitudes to better understand their commuting behaviour and compares predictive models' performances on university commuters' transportation mode choices. For this research, an online survey was administered in March 2020 among the University of Calgary members to shed light on the current travel choice preferences of the university commuters, investigate factors affecting their mode choice, examine their satisfaction level toward various modes, and uncover ways to encourage more sustainable transportation. Further, the data is used to develop traditional discrete choice models and novel machine learning algorithms to predict commuters' transportation mode choice and examine the importance of various factors on their mode choice. The aggregated transportation mode share of the survey respondents shows that 57.89% of survey respondents are either public transit or active mode users, indicating a high percentage of sustainable transportation mode users among university commuters. Various characteristics of survey respondents were shown to be important in their travel behaviour, such as socio-demographic, household and geographical location. It is shown that both age and income level positively affect car usage while negatively affect public transit usage. The geographical analysis also indicates that travel distance and accessibility to transit facilities influence university commuters' mode choice decisions. As travel distance increases and accessibility to transit facilities decreases, university commuters prefer to use more cars than public transit and active modes. Further, it is shown that university members have various satisfaction levels of using different transportation modes and consider various barriers to use them. It is shown that university status and socio-demographics affect commuters' attitudes toward satisfaction level and barriers to use sustainable modes. Overall, employees are shown to be highly concerned about the environmental footprint of using cars; although, they are mainly car commuters. In contrast, students are determined to be more concerned about driving costs and parking availability. In terms of transit, travel time commuting and consistency are determined to be common concerns of university commuters. Similar to transit, the weather condition is determined to be the most important concern for using active modes. The results of the descriptive analysis are further used for policy recommendations to different university administrations to encourage more use of sustainable transportation. Besides descriptive analysis, the prediction performance of various machine learning classifiers is compared with the traditional multinomial logit model on predicting the University of Calgary commuters' transportation modes. The comparison results show that the Extreme Gradient Boosting (XGBoost) method performs better in travel mode choice prediction of the university commuters for higher overall accuracy and F1-score. In addition to the performance comparison, this thesis estimates the relative importance of explanatory variables on various models and shows how they relate to mode choices. This research shows that travel-related information is more influential on machine learning algorithms, while socio-demographic and household characteristics have more effect on the utility functions of the multinomial logit model.
Description
Keywords
Transportation, Travel behaviour, University commuters, Transportation mode choice, Transportation mode choice prediction
Citation
Kamkar, H. (2021). Travel Behavior Analysis and Mode Choice Prediction for Commuting to Campus - Performance Comparison of Discrete Choice and Machine Learning Models (Master's thesis, University of Calgary, Calgary, Canada). Retrieved from https://prism.ucalgary.ca.