A taxonomy of towns : an unsupervised machine learning approach to classify towns in England and Wales

Krenz, Kimon and Penn, Alan and Varoudis, Tasos; (2022) A taxonomy of towns : an unsupervised machine learning approach to classify towns in England and Wales. In: Annual Conference Proceedings of the XXVIII International Seminar on Urban Form. University of Strathclyde Publishing, Glasgow, pp. 906-918. ISBN 9781914241161

[thumbnail of Krenz-etal-ISFU-2021-an-unsupervised-machine-learning-approach-to-classify-towns-in-England-and-Wales]
Text. Filename: Krenz_etal_ISFU_2021_an_unsupervised_machine_learning_approach_to_classify_towns_in_England_and_Wales.pdf
Final Published Version
License: Creative Commons Attribution 4.0 logo

Download (1MB)| Preview


In 2019, the UK Government pledged to focus on levelling up the UK’s underperforming regions. As part of this effort, the Government recognised towns as a vital component in the hierarchical network of regional urban systems. Towns not only link cities to the wider hinterland but can also be places of innovation in their own right. Yet struggling towns, due to their dependence on cities and the surrounding region, often lack the fundamentals to support a strong local economy. Such towns face major socio-economic challenges once economic trajectories decline, including ageing population; lack of existing skills necessary to attract new firms; lack of education; less direct investments; and hindering spatial configurations. Given this context, the aim of this paper is to establish a classification of towns to understand their similarities to support targeted investment and to offer comparative characteristics for policy evaluation. To this end, this study develops a new classification of all towns in England and Wales across a variety of socio-spatial and economic domains. The analysis includes 1,178 urban areas with a population between 5,000 and 225,000. Specifically, we employ 105 workplace-based and residence-based demographic and economic variables of the 2011 Office for National Statistics Census for England and Wales and combine these with newly developed spatial variables on network similarities on the basis of network topology, geometry and centrality metrics. These variables are aggregated on boundaries of the built-up area of towns utilising centroid-based ONS lookup tables. We account for differences in distribution and scale through data transformation and standardisation. We then employ a K-means unsupervised cluster algorithm to establish a two-tier class system, of which the first is presented in this paper. The result is 6 distinctive Supergroups of towns. We further provide descriptive characterisations of each Supergroup and insights into the importance of individual variables.

Persistent Identifier