# RECORD bc5c2c3c86f1f45909dcce75c8f09acff914dd28 # TITLE: Toward transparent and accurate housing price appraisal: Hedonic price models versus machine learning algorithms # YEAR: 2025.0 VENUE: Financial Innovation DOI: 10.1186/s40854-025-00874-w # AUTHORS: S. An; Yena Song; H. Jang; Kwangwon Ahn # PAGES: 29 CANDIDATE TABLE PAGES (1-based): [22, 25] # ABSTRACT: The nonlinearity of hedonic datasets demands flexible automated valuation models to appraise housing prices accurately, and artificial intelligence models have been employed in mass appraisal to this end. However, they have been referred to as “black-box” models owing to difficulties associated with interpretation. In this study, we compared the results of traditional hedonic pricing models with those of machine learning algorithms, e.g., random forest and deep neural network models. Commonly implemented measures, e.g., Gini importance and permutation importance, provide only the magnitude of each explanatory variable’s importance, which results in ambiguous interpretability. To address this issue, we employed the SHapley Additive exPlanation (SHAP) method and explored its effectiveness through comparisons with traditionally explainable measures in hedonic pricing models. The results demonstrated that (1) the random forest model with the SHAP method could be a reliable instrument for appraising housing prices with high accuracy and sufficient interpretability, (2) the interpretable results retrieved from the SHAP method can be consolidated by the support of statistical evidence, and (3) housing characteristics and local amenities are primary contributors in property valuation, which is consistent with the findings of previous studies. Thus, our novel methodological framework and robust findings provide informative insights into the use of machine learning methods in property valuation based on the comparative analysis. ===== PAGE 1 (score 7, numeric-paren 0, table-words 0, var-words 7) ===== Toward transparent and accurate housing price appraisal: Hedonic price models versus machine learning algorithms Sihyun An1, Yena Song2, Hanwool Jang3 and Kwangwon Ahn4*    Introduction Real estate is an essential sector of social and economic systems, and real estate prices fluctuate across time and place, which can affect economic stability (Glaeser et al. 2014; Tchuente and Nyawa 2022). Capturing dynamic patterns in real estate prices requires flexible valuation models that utilize sophisticated algorithms to precisely capture the nonlinearity of housing price datasets. Moreover, housing prices are affected by vari- ous attributes (Anderson and West 2006; Czembrowski and Kronenberg 2016; An et al. 2024); therefore, the interpretability of model outcomes has been accentuated to rein- force our understanding of the pattern in property prices. This approach enables us to Abstract  The nonlinearity of hedonic datasets demands flexible automated valuation models to appraise housing prices accurately, and artificial intelligence models have been employed in mass appraisal to this end. However, they have been referred to as “black- box” models owing to difficulties associated with interpretation. In this study, we compared the results of traditional hedonic pricing models with those of machine learning algorithms, e.g., random forest and deep neural network models. Com- monly implemented measures, e.g., Gini importance and permutation importance, provide only the magnitude of each explanatory variable’s importance, which results in ambiguous interpretability. To address this issue, we employed the SHapley Addi- tive exPlanation (SHAP) method and explored its effectiveness through comparisons with traditionally explainable measures in hedonic pricing models. The results dem- onstrated that (1) the random forest model with the SHAP method could be a reliable instrument for appraising housing prices with high accuracy and sufficient interpreta- bility, (2) the interpretable results retrieved from the SHAP method can be consolidated by the support of statistical evidence, and (3) housing characteristics and local ameni- ties are primary contributors in property valuation, which is consistent with the find- ings of previous studies. Thus, our novel methodological framework and robust find- ings provide informative insights into the use of machine learning methods in property valuation based on the comparative analysis. Keywords:  Hedonic price model, Importance measure, Machine learning, Housing price appraisal Open Access © The Author(s) 2025. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the mate- rial. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://​ creat​iveco​mmons.​org/​licen​ses/​by/4.​0/. RESEARCH An et al. Financial Innovation (2025) 11:141 https://doi.org/10.1186/s40854-025-00874-w Financial Innovation *Correspondence: k.ahn@yonsei.ac.kr 1 Division of Finance and AI, Hankuk University of Foreign Studies, Yongin, South Korea 2 Department of Geography, Chonnam National University, Gwangju, South Korea 3 Department of Finance, Accounting and Risk, Glasgow Caledonian University, Glasgow, UK 4 Department of Industrial Engineering, Yonsei University, Seoul, South Korea ===== PAGE 22 (score 28, numeric-paren 0, table-words 6, var-words 13) ===== Page 22 of 29 An et al. Financial Innovation (2025) 11:141 used to represent the traditional approach. In addition, the RF model was used for this comparative analysis because of its robustness. The benchmark dataset includes housing sale prices as the target variable and com- prises 80 explanatory variables divided into a training set with 1,460 observations and a test set with 1,459 observations. Prior to developing the property valuation models, we conducted exploratory data analysis and applied feature engineering to refine the data- set and extract informative variables. During this process, we observed that some vari- ables had many missing values. For example, the PoolQC variable (indicating the pool quality) had missing values in 99.5% of the total observations, and the MiscFeature vari- able (indicating the miscellaneous feature) was missing in approximately 1,406 of the observations (representing approximately 96.3%). In addition, variables, such as Utilities (type of available utilities), exhibited identical values across 1,459 observations, thereby indicating little variability with a minimal impact on housing price dynamics. Thus, we eliminated these variables from the dataset because they were deemed noise rather than meaningful contributors to appraising housing prices. In addition, we note that there are still missing values in the preprocessed dataset. The benchmark dataset comprises both qualitative and quantitative variables; thus, the miss- ing values were imputed by the missing-indicator method (Zhuchkova and Rotmistrov 2022) for the qualitative variables and the mean imputation (Silva-Ramírez et al. 2011; Khan and Hoque 2020) method for the quantitative variables. The missing-indicator method imputes missing values in a qualitative variable by a numerical value indicating the absence, which has been widely used in previous studies (Trevizo and Lopez 2016; Table 9  Comparisons of two permutation importance measurements The average score is the mean value of the importance score, which is obtained by repeating the measurement of permutation importance that shuffles the data points in variables over 10 iterations Variables Shuffling method (10 iterations) Omitting method OLS SLR OLS SLR Rank Average score Rank Average score Rank Rank Higher degree 1 0.0604 1 0.0599 1 1 Year 2 0.0484 2 0.0488 2 2 Parking 3 0.0354 3 0.0354 3 3 Area 4 0.0213 4 0.0212 4 4 Households 6 0.0098 6 0.0097 5 5 CBD 5 0.0132 5 0.0126 6 6 Floor 7 0.0075 7 0.0074 7 7 Greenspace 8 0.0055 8 0.0056 8 8 Waterfront 9 0.0019 9 0.0019 9 9 Subway 11 0.0010 11 0.0009 10 10 Top school 12 0.0008 12 0.0008 11 11 Spring 10 0.0010 10 0.0010 12 12 Fall 15 0.0004 15 0.0004 13 13 Winter 14 0.0006 13 0.0006 15 15 Pop. density 13 0.0006 14 0.0005 14 14 Heating type 16 0.0002 16 0.0002 16 16 Bus count 17 0.0001 17 0.0001 17 17 ===== PAGE 23 (score 7, numeric-paren 0, table-words 3, var-words 1) ===== Page 23 of 29 An et al. Financial Innovation (2025) 11:141 Weiss et al. 2017). For the quantitative variables, the mean imputation method is a gen- eral approach for handling missing values (Silva-Ramírez et al. 2011) with low compu- tational costs (Khan and Hoque 2020) because it replaces the missing values with the average value for the considered variable. These processes preserve most of the obser- vations across the entire dataset for training valuation models. Finally, the training set contained a total of 1,460 observations with 68 variables. Considering the large number of housing variables in the refined dataset compared with our hedonic dataset, we conducted principal component analysis, which is a gen- eral feature engineering technique in the modeling process that effectively reduces the dimensionality of variable sets (Noori et al. 2010; Wang et al. 2016; Zhang et al. 2018). Here, we considered two sets of principal components determined by two points where explainable variances over 90% and 95% of the total variance were achieved. Figure 6 shows the cumulative explainable variance across varying numbers of principal compo- nents. In particular, 43 principal components explained 90.23% of the variance, and 51 principal components explained 95.31% of the variance. Accordingly, we compared the traditional approach (i.e., the OLS regression model) against the sophisticated RF model in terms of the RMSE value. Three cases were inves- tigated: (1) using the full variable set, (2) using the 43 principal components set, and (3) using the 51 principal components set. Table 10 shows the results obtained on the Kag- gle housing dataset. The RF model consistently achieved lower RMSE values than those of the OLS-based regression model for all the cases. In our analysis, the case without feature engineering exhibited the highest accuracy in the assessment of housing prices. For each case, the RF model demonstrated an average increase in accuracy of 18% com- pared with the OLS-based regression model. These results are well aligned with the Fig. 6  Cumulative explainable variance of principal components Table 10  RMSE results obtained on the Kaggle housing dataset The numbers in parentheses indicate each case: (1) Full variable set, (2) 43 principal components, and (3) 51 principal components OLS-based regression RF (1) (2) (3) (1) (2) (3) RMSE 0.1749 0.1861 0.1896 0.1451 0.1531 0.1532 ===== PAGE 25 (score 20, numeric-paren 0, table-words 8, var-words 2) ===== Page 25 of 29 An et al. Financial Innovation (2025) 11:141 SLRs in terms of both ­R2 and RMSE values because of their ability to mitigate biased estimates (López-Laborda et al. 2021). These results are consistent with those of pre- vious studies (Han and Jang 2013; López-Laborda et al. 2021). However, the RF model consistently demonstrates the highest predictive power in all the cases, emphasizing that RF-based valuation models can be considered a desirable approach for precisely apprais- ing housing prices. Additional accuracy metrics and processing time We use the additional statistical indices, mean absolute error (MAE) and mean absolute percentage error (MAPE), to further assess the predictive power of the models. These indices are commonly employed metrics in model validation (Ho et al. 2021). We train each model using the complete set of hedonic variables to ensure comparability. We also examine the processing time of the ML and DL models, as convergence speed is a key factor in evaluating the computational efficiency of AI-based systems (Yates and Islam 2021; Abed et al. 2022). Table  12 presents the predictive power of the HPMs and AI models. Our findings demonstrate that the RF model consistently outperforms the other models across all the evaluation metrics, including the RMSE, ­R2, MAE, and MAPE. Furthermore, the RF model requires approximately 13% of the processing time of DNN models, underscoring its computational efficiency. These findings confirm the superiority of the RF model in terms of precision and computational efficiency. According to Yates and Islam (2021), the computational complexity of the RF model primarily depends on the number of trees. Correspondingly, a smaller ensemble (i.e., a reduced number of tree estimators) consumes fewer computational resources (Yates and Islam 2021). Nevertheless, the fine-tuning experiment for RF in Table 8 shows that the RF configuration with fewer trees can achieve predictive power comparable to those with a larger number of trees. As such, our resultant measures in Table 12 show that the RF model can be a more efficient tool for housing price valuation with fewer trees while maintaining comparable predictive accuracy. Abbreviations AI Artificial intelligence DL Deep learning DNN Deep neural network GLM Generalized linear model Table 12  Model validation with additional accuracy measures and processing times A lower value of the reported metric indicates that the model has greater predictive power, excluding ­R2. The “Processing time (sec.)” column shows the convergence time of the model training process, measured in seconds. The row “DNNs” indicates the case of the DNN model with min–max scaling RMSE R2 MAE MAPE Processing time (sec.) OLS 0.194 0.653 0.143 0.010 − SLR 0.194 0.651 0.143 0.010 − RF 0.081 0.939 0.053 0.004 14.340 DNN 0.186 0.681 0.137 0.009 112.669 DNNs 0.127 0.850 0.089 0.006 116.708 ===== PAGE 26 (score 11, numeric-paren 0, table-words 0, var-words 11) ===== Page 26 of 29 An et al. Financial Innovation (2025) 11:141 GMSE Generalized mean-squared error HPM Hedonic price model MAE Mean absolute error MAPE Mean absolute percentage error ML Machine learning OLS Ordinary least squares ReLU Rectified linear unit RF Random forest RMSE Root mean square error SHAP SHapley Additive exPlanation SLR Spatial lag regression Acknowledgements The authors would like to thank the editors and the anonymous reviewers for their invaluable comments and suggestions. Author contributions All the authors contributed equally. Funding This work was supported by (i) the National Research Foundation of Korea grant funded by the Korea government (MSIT) (RS-2025-16067531: Kwangwon Ahn) and (ii) Hankuk University of Foreign Studies Research Fund (0f 2025: Sihyun An). Availability of data and materials The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request. Declarations Competing interests The authors declare that they have no competing interests. Received: 7 July 2023 Accepted: 6 November 2025 References Abed M, Imteaz MA, Ahmed AN, Huang YF (2022) Modelling monthly pan evaporation utilising random forest and deep learning algorithms. Sci Rep 12(1):13132 Ahn K, Jang H, Song Y (2020) Economic impacts of being close to subway networks: a case study of Korean metropolitan areas. Res Transp Econ 83:100900 Ali O, Abdelbaki W, Shrestha A, Elbasi E, Alryalat MA, Dwivedi YK (2023) A systematic literature review of artificial intel- ligence in the healthcare sector: benefits, challenges, methodologies, and functionalities. J Innov Knowl 8(1):100333 Alonso Robisco A, Carbó JM (2022) Measuring the model risk-adjusted performance of machine learning algorithms in credit default prediction. Financ Innov 8(1):70 An S, Ahn K, Bae J, Song Y (2024) Economic impacts of a subway system: exploring local contexts in a metropolitan area. Res Transp Bus Manag 56:101188 An S, Jang H, Kim H, Song Y, Ahn K (2023) Assessment of street-level greenness and its association with housing prices in a metropolitan area. Sci Rep 13(1):22577 Anderson ST, West SE (2006) Open space, residential property values, and spatial context. Reg Sci Urban Econ 36(6):773–789 Anselin L (2013) Spatial econometrics: methods and models, vol 4. Springer, Heidelberg Asif MR, Bording TS, Barfod AS, Grombacher DJ, Maurya PK, Christiansen AV, Auken E, Larsen JJ (2021) Effect of data pre- processing on the performance of neural networks for 1-D transient electromagnetic forward modeling. IEEE Access 9:34635–34646 Bailey JR, Lauria D, Lindquist WB, Mittnik S, Rachev ST (2022) Hedonic models of real estate prices: GAM models; environ- mental and sex-offender-proximity factors. J Risk Financ Manag 15(12):601 Bax D, Chasomeris MG (2019) Listing price estimation of apartments: a generalised linear model. J Econ Financ Sci 12(1):1–11 Breiman L (2001) Random forests. Mach Learn 45(1):5–32 Calle ML, Urrea V (2011) Letter to the editor: stability of random forest importance measures. Brief Bioinform 12(1):86–89 Chen Y, Ye Y, Liu X, Yin C, Jones CA (2025) Examining the nonlinear and spatial heterogeneity of housing prices in urban Beijing: an application of GeoShapley. Habitat Int 162:103439 Chien CY, Hsu SW, Lee TL, Sung PS, Lin CC (2020) Using artificial neural network to discriminate Parkinson’s disease from other parkinsonisms by focusing on putamen of dopamine transporter SPECT images. Biomedicines 9(1):12 Chin T, Chau K (2003) A critical review of literature on the hedonic price model. Int J Hous Sci Appl 27(2):145–165 Chuan CH, Sun R, Tian S, Tsai WHS (2024) EXplainable artificial intelligence (XAI) for facilitating recognition of algorithmic bias: an experiment from imposed users’ perspectives. Telemat Inform 91:102135 Czembrowski P, Kronenberg J (2016) Hedonic pricing and different urban green space types and sizes: insights into the discussion on valuing ecosystem services. Landsc Urban Plan 146:11–19