Abstract
The dissertation outlines novel analytical and experimental methods for discrete choice modelling using generative modelling and information theory. It explores the influence of information heterogeneity on large scale datasets using generative modelling. The behaviorally subjective psychometric indicators are replaced with a learning process in an artificial neural network architecture. Part of the dissertation establishes new tools and techniques to model aspects of travel demand and behavioural analysis for the emerging transport and mobility markets. Specically, we consider: (i) What are the strengths, weaknesses and role of generative learning algorithms for behaviour analysis in travel demand modelling? (ii) How to monitor and analyze the identiability and validity of the generative model using Bayesian inference methods? (iii) How to ensure that the methodology is behaviourally consistent? (iv) What is the relationship between the generative learning process and realistic representation of decision making as well as its usefulness in choice modelling? and (v) What are the limitations and assumptions that have needed to develop the generative model systems? This thesis is based on four articles introduced in Chapters 3 to 6. Chapters 3 and 4 introduces a restricted Boltzmann machine learning algorithm for travel behaviour that includes an analysis of modelling discrete choice with and without psychometric indicators. Chapter 5 provides an analysis of information heterogeneity from the perspective of a generative model and how it can extract population taste variation using a Bayesian inference based learning process. One of the most promising applications for generative modelling is for modelling the multiple discretecontinuous data. In Chapter 6, a generative modelling framework is developed to show the process and methodology of capturing higher-order correlation in the data and deriving a process of sampling that can account for the interdependencies between multiple outputs and inputs. A brief background on machine learning principles for discrete choice modelling and newly developed mathematical models and equations related to generative modelling for travel behaviour analysis are provided in the appendices.
Highlights
OverviewThe transport and mobility market is currently undergoing a fundamental transformation, driven by three domains of disruptive technologies: Mobility-as-a-Service (MaaS), Connected and Automated Vehicles (CAV) and Articial Intelligence (AI) [1, 2, 3]
We develop a new approach to the problem of modelling latent behaviour through the estimation of a joint distribution from its associated choice and auxiliary information
The estimation process is straightforward, and convergence is fast for large parameter vectors using stochastic gradient descent with contrastive divergence (CD) objective function
Summary
OverviewThe transport and mobility market is currently undergoing a fundamental transformation, driven by three domains of disruptive technologies: Mobility-as-a-Service (MaaS), Connected and Automated Vehicles (CAV) and Articial Intelligence (AI) [1, 2, 3]. Complex theories of decision-making process provide the basis for latent behaviour representation in statistical models These processes focus on the use of psychometric data, such as choice perception and attitudinal questions. These variations include: decision-protocols, choice sets, unobserved taste variations and unobserved attributes [26] Under these considerations, recent studies on travel behaviour analysis have so far primarily focused on representing heterogeneity in the error correction function and incorporating it into utility based multinomial logit (MNL) models [108]. The standard multinomial logit (MNL) model used in travel behaviour modelling relies on deterministic decision rules utilities are operationalized by random variables and it is assumed that unobserved heterogeneities are independent and identically distributed (i.i.d.) [21]. Flexible error component structural models have been developed to relax the independence of irrelevant alternatives (i.i.a.) assumption by parameterizing an appropriate error variance as a function of individual attributes
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
