Evaluating travel mode decisions and transport models in understanding transit equity: The case of greater Toronto and Hamilton area

Abstract

In recent decades, the incorporation of equity considerations in the transportation domain and the equity analysis of transport projects and policies are rapidly increasing. These approaches mainly include travel behaviour analysis with equity indicators and the socioeconomic impacts of transport investments on individuals. Accordingly, the cost and benefits of transport investments for residents are evaluated. Moreover, travellers' travel behaviour, daily activity patterns, and travel mode decisions are estimated through their trip chains analysis. These assessments can offer a broad perspective on individuals' travel needs and constraints. They also offer valuable insight for transportation planners and policymakers in understanding how different transport investments impact society. Therefore, they enable authorities and planners to develop equitable transport policies and travel demand management to address various environmental problems. This dissertation focuses on understanding how different socioeconomic groups plan their daily trips and reports important findings on their responses to transport investments, aiming to improve individuals' activity participation and alleviate travel barriers. The study also evaluates travel behaviour and mode use models and investigates the potential of machine learning algorithms for travel behaviour prediction in the Greater Toronto and Hamilton Area (GTHA), one of the largest and fastest-growing regions in Canada. The primary data source used for this study is the 2016 Transportation Tomorrow Survey (TTS) dataset, a large-sample household travel survey including a one-day household travel diary conducted in the Greater Golden Horseshoe Area. The TTS data is a part of an ongoing data collection program started in 1986 and is collected every five years. This regional survey is conducted to travel demand management, and it can use for transportation planning programs and models. In the first step, this study explores how income and car-ownership levels determine activity patterns and travel decisions of travellers using an aggregated form of activity type and travel mode as a unit of trip chain analysis. A presumption-free clustering framework is leveraged to mitigate the subjectivity of rule-based approaches for trip chain analysis. This approach extracts the homogeneous clusters of activity patterns. Second, the impacts of transit improvements in low-income communities are explored based on the assumption that transit investments could result in changing travel mode use and generating more transit and fewer car trips. Such analysis is performed by exploring the association between transit use and transit accessibility improvements using stratified regression models. Lastly, the effects of travel behaviour models are evaluated in terms of their predictive performance in policy-making and transportation planning. This study investigates how the model selection affects the prediction of transit use and compares the predictive performance of traditional and Machine Learning (ML) algorithms. Then, it evaluates a transit investment policy by contrasting the predicted activities and the spatial distribution of transit trips generated by the vulnerable households after improving accessibility. The findings of this study reveal that income and car-ownership levels influence a traveller's travel decisions and change their mobility patterns. The findings show that females, regardless of income or car ownership, frequently take transit in their daily trip chains. Among low-income carless individuals, most of their daily trips include the mobility of care, where women more often than men play this traditional role in a household by either public transit or a car as a passenger. In the low-income car-owner subsample, females still use public transit for their work trips, whereas males regularly use the household's car to commute to work. It confirms that women benefit less from having access to a car in families with a shared private vehicle. Males of wealthy carless households integrate public transit and active transportation for their daily trips when they live in high-density and more accessible neighbourhoods. Furthermore, evaluating transit improvements in low-income communities shows that low-income households with one or more cars per adult have the most elastic relationship between transit accessibility and transit use; they are more likely to be transit riders if transit improves. However, in auto-centric areas with poor transit, the transit use of low-income households drops off sharply as car ownership increases. It implies that low-income car-owning households might become too reliant on their vehicles as soon as they own them. Moreover, the sensitivity analysis exploring how changes to accessibility affect transit trip generation highlights that the accessibility gains in the region provide more opportunity for increasing transit ridership among car-deficit households when transit is improved. Therefore, the analysis suggests some insight into engaging individuals in taking transit and resulting in overall transit ridership in the region. Given the model selection, the results show that ML algorithms outperform all other statistical models and have great potential for enhancing travel behaviour predictions without sacrificing interpretability. Random Forest (RF), XGBoost (XGB), and Neural Networks (NN) classifiers and regressors significantly outperform other algorithms. Among them, RF is the most accurate approach for predicting low-income families' transit demand according to its predictive performance. However, statistical models perform poorly when forecasting transit users' behaviours. Further, the spatial distribution of newly generated transit trips after transit improvements is not identical; thus, traditional models may arrive at a different, probably inaccurate, policy recommendation in addressing social, spatial, and environmental problems. Moreover, applying model-agnostic interpretation tools to ML models shows that these techniques can uncover each model's underlying process, which was supposed to be a "black box". All in all, ML models demonstrate significant improvement in accuracy and interpretability. The findings point out that understanding and estimating individuals' travel decisions and preferences with a reliable model enables policymakers to establish an appropriate transit framework that benefits low-income people and alleviates transit inequality in society. This study suggests that evaluating individuals' travel behaviour in terms of their income and car-ownership levels may give a new and different outlook on transport planning in metropolitan cities. Overall, a fair transportation investment that meets environmental, economic, and social goals necessitates a thorough understanding of different socioeconomic groups' travel requirements and responses. The findings help planners rethink transport policies and strategies that increase activity participation and reduce environmental impacts.