UCL Discovery
UCL home » Library Services » Electronic resources » UCL Discovery

Spatial distribution of building use: recognition and prediction of use with machine learning

Thirapongphaiboon, T; Hanna, S; (2019) Spatial distribution of building use: recognition and prediction of use with machine learning. In: The 12th International Space Syntax Symposium. International Space Syntax Symposium: Beijing, China. Green open access

[thumbnail of ThirapongphaiboonHanna.pdf] Text
ThirapongphaiboonHanna.pdf - Published Version

Download (3MB)

Abstract

Spatial measures that can be applied at different scales, such as angular choice at specific radii, have been shown to effectively identify distinct and important spatial patterns at each scale. Foreground movement routes, local commercial centres, and identifiable urban regions have each be associated with different scales of such measures. This paper investigates and quantifies the degree to which the likelihood of use types of individual buildings can be determined based on spatial measures of the street segment graph alone, using supervised machine learning on a detailed dataset of buildings in London. A vector of 66 dimensions representing the spatial profile of each street segment is taken from analysis in DepthmapX and includes graph centrality measures such as choice at various radii, along with immediate features such as segment length. The proportion of building use is taken from OpenStreetMap and presented as a proportion of buildings in each category of Residential, Commercial or Business. A multilayer perceptron with between one and three hidden layers is trained to output the expected proportions of building use category given the various-dimensional spatial input vector for each segment. Various configurations of the training set and the network configuration are tested and discussed. Results of training indicate a best accuracy of approximately 85% with correlation coefficient of 0.696 for training set and 0.517 (moderately strong) for test set with 37-dimensional spatial input vector, suggesting that the spatial predictability of building use is comparable to that of other human factors such as movement. The spatial factors that appear most significant to the distribution of use types are variations in metric total length and metric mean depth SLW at medium metric radii. An analysis of the most significant dimensions of the input is given, to suggest a spectral spatial profile of each building use type, and the effectiveness of these in prediction or planning is discussed.

Type: Proceedings paper
Title: Spatial distribution of building use: recognition and prediction of use with machine learning
Event: The 12th International Space Syntax Symposium
Location: Beijing, China
Dates: 08 July 2019 - 13 July 2019
Open access status: An open access version is available from UCL Discovery
Publisher version: http://www.12sssbeijing.com/proceedings/download.p...
Language: English
Additional information: This version is the version of record. For information on re-use, please refer to the publisher’s terms and conditions.
Keywords: Neural Networks, Land Use, Typology
UCL classification: UCL
UCL > Provost and Vice Provost Offices > UCL BEAMS
UCL > Provost and Vice Provost Offices > UCL BEAMS > Faculty of the Built Environment
UCL > Provost and Vice Provost Offices > UCL BEAMS > Faculty of the Built Environment > The Bartlett School of Architecture
URI: https://discovery.ucl.ac.uk/id/eprint/10080662
Downloads since deposit
274Downloads
Download activity - last month
Download activity - last 12 months
Downloads by country - last 12 months

Archive Staff Only

View Item View Item