ALF-Score: a predictive, personalized, transferable and network-based walkability scoring system

Alfosool, Ali M. S. (2021) ALF-Score: a predictive, personalized, transferable and network-based walkability scoring system. Doctoral (PhD) thesis, Memorial University of Newfoundland.

[img] [English] PDF - Accepted Version
Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission.

Download (52MB)

Abstract

Measuring the environments around us, including cities, roads, and social environments is crucial to understanding human behaviour. This knowledge can help us predict how aspects of the environment influence behaviour and health. Walkability is a popular measure of the environment used to describe various aspects of the built and social environment associated with physical activity and public health. Most existing methods are missing or underutilizing some crucial parameters that substantially impact measuring accurate walkability scores. For instance, road network structure is an integral part of mobility and should be an essential part of walkability but is not widely discussed in existing methods. Moreover, most walkability measures provide area-based walkability, or their scores are distributed with low spatial resolution. Additionally, individuals’ opinions are not considered when measuring walkability. Furthermore, walkability is subjective, and although multiple definitions of walkability exist, there is no single agreed-upon definition. Existing measures take a one-size-fits-all approach without providing any personalization based on users’ perspectives, leaving much more desired. In this research, Active Living Feature Score or ALF-Score¹ is proposed, which is a novel approach to measure walkability scores more accurately and efficiently while addressing existing limitations. ALF-Score incorporates road network structure to derive features such as network science centralities and network embedding, which are crucial in understanding the road structure better. ALF-Score utilizes user opinion to build a high-confidence ground-truth used to generate models capable of estimating walkability scores based on user opinion. By incorporating machine learning approaches in its pipelines, ALF-Score achieved a much higher granularity and higher spatial resolution of walkability scores at point level. ALF-Score introduced two new methods: 1) a combined graph reduction and reconstruction technique that focuses on reducing the number of nodes in a road network achieving an average of 77% reduction while preserving the core structure of the road network, and 2) the Generalized Linear Extension of Partial Orders or GLEPO, which enables the conversion of relative rankings to absolute scores. Moreover, ALF-Score+ extends ALF-Score by incorporating user demographics such as age and gender to capture profile clusters that help provide personalized walkability scores suitable for varying individual profiles. Additionally, ALF-Score++ improves the overall scalability of ALF-Score and further extends this measure by incorporating transferability to allow reusability of already-learned knowledge and previously detected patterns as a base for further and continued learning to help reduce training time, improve prediction accuracy, reduce resource consumption, and lower the number of labels needed for training. Most importantly, ALF-Score++ enables zero-user- input application, which allows predicting walkability scores for any location on the road without training models for that particular region with a low transferability loss of 13.28 units using Deep Neural Network approaches, and a direct training loss of only 4.56 units using shallow learners (MAE on a scale of 0-100). ¹https://alfscore.com/

Item Type: Thesis (Doctoral (PhD))
URI: http://research.library.mun.ca/id/eprint/15347
Item ID: 15347
Additional Information: Includes bibliographical references (pages 210-228).
Keywords: walkability, complex networks, machine learning, public health, environmental policy, road networks, personalized walkability, transferable walkability, ALF-Score
Department(s): Science, Faculty of > Computer Science
Date: December 2021
Date Type: Submission
Digital Object Identifier (DOI): https://doi.org/10.48336/98YN-PJ37
Library of Congress Subject Heading: Walking--Environmental aspets; Walking--Social aspects; Machine learning.

Actions (login required)

View Item View Item

Downloads

Downloads per month over the past year

View more statistics