Back to search

Democritus University of Thrace (DUTH)

Applications of artificial intelligence in transport planning discrete choice modelling

Abstract

dc:description

The present thesis regards to a research program which aimed to develop and test machine learning techniques for travel demand forecasting, especially discrete choice modelling, develop and evaluate performance metrics that can be applied to models developed with different techniques, and outline a framework for the incorporation of machine learning models in state-of-the-art transport planning models. A number of different models based on very different kinds of datasets were developed and a variety of different performance metrics were tested on them. It was found that machine learning techniques performed favorably compared to more conventional counterparts, both for regression and classification tasks, and that with modern workstation hardware necessary training times are limited with the exception of the Adaptive Neuro-Fuzzy Inference System, which required an extremely large amount of time for applications with moderate to large dimensionality. A number of different performance metrics were tested, which included measures to detect overfitting in trained models and poor performance over minority classes. Based on already existing metrics, such as the F1-score, as well as the aforementioned proposed metrics, multicriteria scores were developed to compare models developed with different techniques that also account for model complexity by considering the number of explanatory variables or the size of the hidden layer. These metrics pointed to better performance for machine learning models over their Logit counterparts, both multinomial and nested ones. It was found that severely unbalanced datasets, which is a common case for travel planning study surveys where private cars are often over-represented, may lead to severe cases of overfitting. This can be overcome by employing undersampling, where majority classes are under-represented in the training dataset compared to the sample (e.g., the results of a transport planning survey), oversampling, where minority classes are over-represented via the introduction of properly prepared synthetic datapoints, or hybrid techniques which combine these two approaches. Incorporating individuals’ perceptions of different options (e.g., regarding their cost or the social status they conveyed to their user) proved to be beneficial and to improve the performance of discrete choice models compared to ones developed with the same techniques that did not include these. Two approaches were tried to achieve this goal; the first one was to treat respondents’ perceptions as linear ordinal variables and incorporate them in a non-fuzzy machine learning model, specifically Multilayer Perceptrons, Support Vector Machines, or Radial Basis Function Networks, whereas the second one was to use an Adaptive Neuro-Fuzzy Inference System and consider them as fuzzy input variables. The second approach proved to be unwieldy, in the sense that it requires very long training times compared to its aforementioned non-fuzzy counterparts, while offering little to no performance gains. It may be possible that a different approach to incorporate the fuzziness of individuals’ responses regarding their perceptions of different choices’ characteristics is more productive and it is suggested that future research explore this utilizing some of the model comparison metrics developed for the present study. Chapter 1 provides an introduction to the problem while Chapter 2 discusses the theoretical background of transport planning and statistical pattern recognition with an emphasis on Machine Learning. Chapter 3 presents the literature review that was undertaken, whereas Chapter 4 discusses the methodology that was followed during the present program’s experiments for regression and classification tasks. Chapter 5 presents the results of the experiments, while Chapter 6 discusses their practical implications with regard to the application of Machine Learning techniques in transport planning. Finally, Chapter 7 provides a brief overview of the present research program’s conclusions.

Degree

thesis:*
Grantor dc:publisher
Democritus University of Thrace (DUTH)
Year dc:date
2024

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Christidis, Konstantinos
  • Χρηστίδης, Κωνσταντίνος

Subjects

dc:subject × 16

Rights

Language dc:language
eng

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/58487

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Christidis, Konstantinos; Χρηστίδης, Κωνσταντίνος. Applications of artificial intelligence in transport planning discrete choice modelling. Democritus University of Thrace (DUTH), 2024. http://hdl.handle.net/10442/hedi/58487