The Library
Using deep learning to infer house prices from street view imagery
Tools
Chahal, Bhavan Kaur (2022) Using deep learning to infer house prices from street view imagery. PhD thesis, University of Warwick.
|
PDF
WRAP_Theses_Chahal_2022.pdf - Unspecified Version - Requires a PDF viewer. Download (70Mb) | Preview |
Official URL: http://webcat.warwick.ac.uk/record=b3912179~S15
Abstract
House prices are a key economic indicator. However, there is usually a delay of two months between the sale of a house and the availability of price data through the Land Registry. Over the past decade, vast quantities of imagery of streets and neighbourhoods have become available online, and the speed at which such imagery is being updated is growing. Simultaneously, advances in deep learning have enhanced our ability to determine the contents of an image automatically. In this thesis, we therefore ask: can we use deep learning to automatically infer house prices from large volumes of photos of houses and neighbourhoods?
Focusing on London as a case study, we analyse millions of photos from three sources, covering photographs of streets (Google Street View); crowdsourced photographs of outdoor neighbourhood locations (Geograph); and marketing images of house exteriors and interiors (Zoopla). For each image, we use a convolutional neural network to extract data on the presence of visual features. We investigate whether these features can be used in conjunction with an elastic net regression model to estimate the median house price during 2015 and 2016 for London neighbourhoods.
Our results demonstrate that it is possible to automatically infer local house prices from all three sources of imagery. The best estimates are made using Zoopla photographs. This may be due to the larger number of Zoopla images available in our sample (3,727,890 photographs, compared to 519,295 images for Google Street View and 273,999 images for Geograph). However, we find that increasing the number of images available to the model by combining images from different sources has limited benefits.
As the speed of collection of such imagery increases, this approach could equip national and local policymakers with more timely information on house prices, enabling them to make more informed decisions about economic policy.
Item Type: | Thesis (PhD) | ||||
---|---|---|---|---|---|
Subjects: | H Social Sciences > HD Industries. Land use. Labor T Technology > TA Engineering (General). Civil engineering (General) |
||||
Library of Congress Subject Headings (LCSH): | Housing -- Prices -- England -- London, Housing -- Prices -- England -- London -- Statistical methods, Image analysis, Image processing -- Digital techniques, Deep learning (Machine learning) | ||||
Official Date: | September 2022 | ||||
Dates: |
|
||||
Institution: | University of Warwick | ||||
Theses Department: | Mathematics for Real-World Systems Centre for Doctoral Training | ||||
Thesis Type: | PhD | ||||
Publication Status: | Unpublished | ||||
Supervisor(s)/Advisor: | Moat, Suzy ; Preis, Toby | ||||
Sponsors: | Engineering and Physical Sciences Research Council ; Medical Research Council (Great Britain) | ||||
Format of File: | |||||
Extent: | xii, 181 leaves : colour illustrations, colour maps | ||||
Language: | eng |
Request changes or add full text files to a record
Repository staff actions (login required)
View Item |