1# RECORD 2cdb6f374417590715ab5db24f593d17616b9ea22# TITLE: Effects of neighborhood streetscape on the single-family housing price: Focusing on nonlinear and interaction effects using interpretable machine learning3# YEAR: 2025.0 VENUE: PLoS ONE DOI: 10.1371/journal.pone.03234954# AUTHORS: J. Han; Ayoung Woo; Sugie Lee5# PAGES: 25 CANDIDATE TABLE PAGES (1-based): [13]6# ABSTRACT: Previous studies using the conventional Hedonic Price Model to predict existing housing prices may have limitations in addressing the relationship between house prices and streetscapes as visually perceived at the human eye level, due to the constraints of streetscape estimations. Therefore, in this study, we analyzed the relationship between streetscapes visually perceived at eye level and single-family home prices in Seoul, Korea, using computer vision technology and machine learning algorithms. We used transaction data for 13,776 single-family housing sales between 2017 and 2019. To measure visually perceived streetscapes, this study used the Deeplab V3 + deep-learning model with 233,106 Google Street View panoramic images. Then, the best machine-learning model was selected by comparing the explanatory powers of the hedonic price model and all alternative machine-learning models. According to the results, the Gradient Boost model, a representative ensemble machine learning model, performed better than XGBoost, Random Forest, and Linear Regression models in predicting single-family house prices. In addition, this study used an interpretable machine learning model of the SHAP method to identify key features that affect single-family home price prediction. This solves the “black box” problem of machine learning models. Finally, by analyzing the nonlinear relationship and interaction effects between perceived streetscape characteristics and house prices, we easily and quickly identified the relationship between variables the hedonic price model partially considers.789===== PAGE 1 (score 6, numeric-paren 0, table-words 0, var-words 6) =====1011PLOS One | https://doi.org/10.1371/journal.pone.0323495 May 21, 2025121 / 2513 14 OPEN ACCESS15Citation: Han J, Woo A, Lee S (2025) Effects of 16neighborhood streetscape on the single-family 17housing price: Focusing on nonlinear and 18interaction effects using interpretable machine 19learning. PLoS One 20(5): e0323495. https://20doi.org/10.1371/journal.pone.032349521Editor: Zeyar Aung, Khalifa University, UNITED 22ARAB EMIRATES23Received: June 18, 202424Accepted: April 8, 202525Published: May 21, 202526Copyright: © 2025 Han et al. This is an open 27access article distributed under the terms of 28the Creative Commons Attribution License, 29which permits unrestricted use, distribution, 30and reproduction in any medium, provided the 31original author and source are credited.32Data availability statement: Please note 33that our data is available at the following 34link (Figshare) https://doi.org/10.6084/35m9.figshare.26965252.36Funding: This work was supported by the 37National Research Foundation of Korea (NRF) 38grant funded by the Korea government (*MSIT) 39RESEARCH ARTICLE40Effects of neighborhood streetscape on 41the single-family housing price: Focusing 42on nonlinear and interaction effects using 43interpretable machine learning44Jaewon Han1, Ayoung Woo452, Sugie Lee461*471 Department of Urban Planning & Engineering, Hanyang University, Seoul, South Korea, 2 Graduate 48School of Urban Studies, Hanyang University, Seoul, South Korea 49* sugielee@hanyang.ac.kr50Abstract 51Previous studies using the conventional Hedonic Price Model to predict existing 52housing prices may have limitations in addressing the relationship between house 53prices and streetscapes as visually perceived at the human eye level, due to the 54constraints of streetscape estimations. Therefore, in this study, we analyzed the 55relationship between streetscapes visually perceived at eye level and single-family 56home prices in Seoul, Korea, using computer vision technology and machine learning 57algorithms. We used transaction data for 13,776 single-family housing sales between 582017 and 2019. To measure visually perceived streetscapes, this study used the 59Deeplab V3 + deep-learning model with 233,106 Google Street View panoramic 60images. Then, the best machine-learning model was selected by comparing the 61explanatory powers of the hedonic price model and all alternative machine-learning 62models. According to the results, the Gradient Boost model, a representative ensem63ble machine learning model, performed better than XGBoost, Random Forest, and 64Linear Regression models in predicting single-family house prices. In addition, this 65study used an interpretable machine learning model of the SHAP method to identify 66key features that affect single-family home price prediction. This solves the “black 67box” problem of machine learning models. Finally, by analyzing the nonlinear relation68ship and interaction effects between perceived streetscape characteristics and house 69prices, we easily and quickly identified the relationship between variables the hedonic 70price model partially considers.711. Introduction72The estimation of housing prices has been a longstanding topic for urban planners 73and policymakers, as housing units are one of the most expensive commodities in 74household finances. Housing prices are closely related to macroeconomic conditions, 757677===== PAGE 13 (score 32, numeric-paren 0, table-words 4, var-words 17) =====7879PLOS One | https://doi.org/10.1371/journal.pone.0323495 May 21, 20258013 / 2581Second, latitude (north-south) and longitude (east-west), which represent the location properties of the houses, had 82negative and positive effects on the predicted price of single-family houses, respectively. The polarization of the housing 83supply within Seoul and the differences in urban infrastructure are severe, and those differences are reflected in housing 84prices. In addition, in the local Shapley value graph, the red and blue dots are mixed, which indicates that a non-linear 85relationship exists between location (latitude and longitude) and single-family house prices.86Table 1. Definitions of the variables and basic statistics.87Variables88Measurements89Mean90S.D.91Min.92Max.93Dependent variable94 ln (housing prices)95The natural logarithm of the price of a single-family house (KRW)9620.17970.559817.329921.24100Independent variables101Structural features102 Site area103The site area of a single-family house (m2)104140.331051,875.981063.00107220,220108 Gross floor area109The total floor area of a single-family house (m2)110122.6311170.211124.50113479.94114 Housing age115Age of the building11639.7011711.06118011995120 Building structure:121Ferroconcrete122Ferroconcrete = 1, others = 01230.011240.1212501261127 Building structure:128Masonry129Masonry = 1, others = 01301.46e-031310.0413201331134 Land topography135Flatland = 1, others = 01360.671370.4713801391140 Site shape141Regular form = 1, others = 01420.841430.3714401451146 Corner parcel147Corner parcel = 1, others = 01480.131490.3315001511152 Parking availability153On-site parking allowed = 1, on-site parking not allowed = 01540.011550.1215601571158Walking environment159Macro-scale walking environment factors160 3-way intersection161density162(Number of 3-way intersections/total street length) within the 163400m buffer1643.29E-031651.59e-0316601670.02168 4-or-more-way169intersection density170(Number of 4-or-more-way intersections/total street length) within 171the buffer1726.85e-041735.34e-0417401753.51e-03176 Subway station density177(Number of subway stations/total street length) within the buffer1785.73e-041799.55e-0418001810.01182 Bus stop density183(Number of bus stops/total street length) within the buffer1840.011850.0518601873.56188 Crosswalk density189(Number of crosswalks/total street length) within the buffer1900.021910.0619201937.34194 Park land use195Total park and open space area within the buffer (m2)19644,431.2419766,091.751980199447,553.80200 Residential land use201Total area for residential use within the buffer (m2)202441,246.40203183,936.502041,615.142051,421,861.00206 Business land use207Total area for businesses within the buffer (m2)20819,325.6220940,873.862100211917,400.30212 Commercial land use213Total area for commercial use within the buffer (m2)21473,537.9321554,375.072160217681,313.20218 Mixed land use219–220k221∑222i=1223(pi) ∗ln(pi)224ln(k)225pi=Total area of land use in the 400m buffer2260.152270.022280.972290.97230k=Number of land uses231Streetscape environment232 ln (greenery)233The natural logarithm of the average greenery value within the buffer2341.632350.462360.182373.93238 ln (enclosure)239The natural logarithm of the average enclosure value within the 240buffer2414.692420.222431.932445.14245 ln (paved sidewalks)246The natural logarithm of the average paved sidewalk value within 247the buffer2482.522490.122502.012513.13252Locational characteristics253 ln (distance to parks)254The natural logarithm of the distance to the nearest park 255boundary (m)2564.862571.0025802596.95260 Distance to CBD261The shortest distance to the CBD boundary (m)2623,765.002632,242.81264026511,539.12266https://doi.org/10.1371/journal.pone.0323495.t001267268269===== PAGE 14 (score 18, numeric-paren 0, table-words 2, var-words 13) =====270271PLOS One | https://doi.org/10.1371/journal.pone.0323495 May 21, 202527214 / 25273Third, the distance from the city center and the green area around each single-family house, which also represent 274location characteristics, had a negative effect on the predicted price of single-family houses. The CBD and parks are 275perceived negatively in a residential environment because they draw many people from outside the neighborhood. On the 276other hand, the business district area, commercial district area, and subway station density had positive effects because 277the various benefits provided by the commercial and business facilities that concentrate around subway stations. In the 278local Shapley value graph, the red and blue dots are mixed, indicating a non-linear relationship between the location and 279environment around a neighborhood and single-family house prices within it.280Fourth, bus stops and land use mix, which are closely related to walkability, had negative and positive effects on hous281ing prices, respectively. Many people visit a bus stop, which are often near busy streets and experience various pollutants 282Table 2. Performance comparison of machine learning models.283Model284Performance evaluation indicator285R2286MSE287MAE288MedAE289Linear Regression2900.6602910.1042920.2482930.197294Random Forest2950.3182960.2092970.3632980.318299XGBoost3000.7653010.0723020.2003030.155304Gradient Boost3050.8173060.0563070.1693080.122309https://doi.org/10.1371/journal.pone.0323495.t002310Fig 6. Analysis of variable importance to single-family housing prices using the SHAP method.311https://doi.org/10.1371/journal.pone.0323495.g006312