1Review23Comparative Analysis of Advanced Models for Predicting4Housing Prices: A Review5Inmaculada Moreno-Foronda 1 , María-Teresa Sánchez-Martínez 1 and Montserrat Pareja-Eastaway 2, *67 1 Department of Applied Economics, University of Granada, 18011 Granada, Spain;8 inmamf@correo.ugr.es (I.M.-F.); tsanchez@ugr.es (M.-T.S.-M.)9 2 Department of Economics, University of Barcelona, 08034 Barcelona, Spain10 * Correspondence: mpareja@ub.edu111213 Abstract: Understanding the determinants of housing price movements is an ongoing14 subject of debate. Estimating these determinants becomes a valuable tool for predicting15 price trends and mitigating the risks of market volatility. This article presents a systematic16 review analyzing studies that compare various machine learning (ML) tools with hedonic17 regression, aiming to assess whether real estate price predictions based on mathematical18 techniques and artificial intelligence enhance the accuracy of hedonic price models used19 for valuing residential properties. ML models (neural networks, decision trees, random20 forests, among others) provide high predictive capacity and greater explanatory power due21 to the better fit of their statistical measures. However, hedonic regression models, while22 less precise, are more robust, as they can identify the housing attributes that most influence23 price levels. These attributes include the property’s location, its internal features, and the24 distance from the property to city centers.2526 Keywords: machine learning; hedonic prices; prediction; prices; housing2728293031 1. Introduction32 Real estate assets serve a dual purpose: they are an essential consumer good, and their33Academic Editor: Maria acquisition represents the most significant investment in the portfolios of most households.34Rosaria Guarini35 Therefore, a thorough understanding of housing price predictability is essential for making36Received: 23 December 2024 informed investment decisions and ensuring the efficient allocation of assets.37Revised: 23 January 202538 Predicting housing prices is crucial not only for households and other market partici-39Accepted: 29 January 202540 pants but also for policy design, economic research, and the broader economy, given the41Published: 31 January 202542 macroeconomic significance of residential price trends [1]. Sustained increases in housing43Citation: Moreno-Foronda, I.;44 prices pose a serious problem in virtually all countries, as they make it difficult for many45Sánchez-Martínez, M.-T.;46Pareja-Eastaway, M. Comparative47 citizens to access housing.48Analysis of Advanced Models for However, predicting housing prices is complex due to the significant challenges in49Predicting Housing Prices: A Review. measuring all the intrinsic and extrinsic characteristics that form part of a property [2].50Urban Sci. 2025, 9, 32. https:// Among the different variables/factors that influence housing prices are, on the one hand,51doi.org/10.3390/urbansci902003252 the internal variables or micro factors, which refer to the structural characteristics of53Copyright: © 2025 by the authors. the property itself. These include specific features of the property, such as size, number54Licensee MDPI, Basel, Switzerland. of rooms, or type of construction, as well as specific characteristics of the building and55This article is an open access article56 common areas that are part of it (elevator, natural gas installations, swimming pool, or57distributed under the terms and58 park, among others). On the other hand, there are external variables or macro factors59conditions of the Creative Commons60Attribution (CC BY) license61 that encompass all types of location and neighborhood attributes of the property, among62(https://creativecommons.org/ which accessibility and distance to essential services, the economic and social conditions of63licenses/by/4.0/). the district where it is located, or the particularities of urbanization of the surroundings,646566Urban Sci. 2025, 9, 32 https://doi.org/10.3390/urbansci902003267 Urban Sci. 2025, 9, 32 2 of 2068697071 such as sidewalks, lighting, etc., stand out [3]. Finally, and no less important, are the72 regional factors, which refer to the exact location where the property is located, including73 transportation conditions, public support facilities, commercial prosperity, etc., as well as74 the socio-economic characteristics associated with the geographic location, although the75 latter are difficult to quantify [4].76 There are two general approaches to evaluating prices in the real estate market. The77 first is the traditional method, where the value of a property is estimated through the78 appraisal or opinion of a trained and experienced expert in the field. This method employs79 comparison, cost, accounting, development, and capitalization methods, among others,80 which inevitably retain a high degree of subjectivity [5].81 The second approach involves advanced prediction models, such as the hedonic82 price model (HPM), machine learning models, case-based reasoning, and spatial analysis83 methods.84 Among advanced price prediction models, hedonic price models (HPMs) stand out.85 HPMs have been more widely adopted by academics and professionals for property86 valuation in various real estate markets worldwide. These models measure the contribution87 of property attributes as well as other external factors that could influence the value of a88 property [6].89 However, over time, and with the continuous improvement of technologies and90 mathematical and statistical procedures, new advanced property valuation methods have91 been developed, offering results that are closer to reality. The use of artificial intelligence92 enables the creation of fully objective prediction models, such as artificial neural network93 models, random forest models, or support vector machines, among others [7].94 In recent decades, advancements in technology and statistical procedures have enabled95 the acquisition and storage of larger volumes of data. Big data applications are attracting96 growing interest among urban researchers. Notable developments include advances in97 data engineering and the use of machine learning techniques for applications across nearly98 all economic sectors. Machine learning is a branch of artificial intelligence, emerged with99 the aim of addressing the issue of subjectivity when determining housing prices [8]. It is100 one of the most widely used methods for predicting housing price models, through which101 computers identify patterns in data without being explicitly programmed for that task [9].102 This methodology imitates human reasoning, and due to its problem-solving capabilities,103 the models it encompasses are efficient alternatives for housing price prediction, as they can104 identify non-linear relationships among relevant data. The development of these models105 can significantly aid in forecasting future housing prices and in establishing real estate106 policies [10,11].107 The use of this tool, which does not rely on socioeconomic behavior models, raises108 doubts about its effectiveness in providing better predictions than traditional methodolo-109 gies [12].110 Despite growing concerns about the use of new prediction models in the real estate111 sector, there are no systematic studies identifying whether these pricing measurement112 methodologies are actively used in the real estate sector. Nor is there a review of which113 models are most widely used, which provide the most accurate results in predicting housing114 prices, in which countries these studies are conducted, or which of them are carried out115 with real data and for experimental purposes. Furthermore, despite the increasing use of116 machine learning models for housing price prediction, there are no studies that compare117 several predictive models for real estate in a specific geographic area.118 The general aim of this paper is to address this gap in the literature, along with a119 series of specific objectives: First, a systematic review is conducted to understand the120 different advanced models used for housing price prediction and to determine in which121 Urban Sci. 2025, 9, 32 3 of 20122123124125 countries these models are most frequently used, what methods experts employ, and which126 models provide the most significant results when predicting real estate prices. Second,127 the paper seeks to compare several advanced prediction models and their consideration128 of the different internal and external characteristics that influence the establishment of129 a property’s value. Finally, the paper aims to present the results of comparing different130 methods applied to real cases in the housing market across different international regions131 for housing suppliers and institutional entities when formulating their economic policies.132 The structure of the article is as follows: the next section provides an overview of133 the literature on advanced real estate valuation and prediction models, followed by a134 description of the methodology employed for the systematic review of the identified135 articles. Lastly, an analysis and discussion of the results are presented.136137 2. Advanced Housing Price Prediction Methods: A Classification138 This section presents a classification of different advanced approaches for real estate139 price valuation: HPMs, neural networks, spatial analysis, etc.140 Hedonic pricing models (HPMs) date back to 1939 when Andrew Court succeeded141 in estimating an index for the automotive market using this model [13]. Technically, the142 theory behind HPMs originates from Lancaster’s renowned consumer theory and Rosen’s143 theoretical model [14]. Rosen stood out for being the first to apply HPMs within standard144 economic theory, incorporating properties considered heterogeneous in nature. Over the145 past decades, this model has been widely used in housing market research due to the high146 heterogeneity of housing characteristics [15].147 HPMs aim to establish, through implicit price equations, the value of a property based148 on its attributes including structural attributes (quality, age, and size of the building),149 location attributes (proximity to transportation links, employment centers, and services),150 and neighborhood attributes (crime levels and socioeconomic profile) [15]. This approach151 explains its relevance in real estate market studies by considering housing as a hetero-152 geneous good composed of various features. The hedonic methodology is considered a153 viable way to study the real estate market, allowing for the estimation of the values of154 each housing attribute that buyers and sellers consider during the transaction process—an155 aspect that can be highly useful to stakeholders [16].156 Artificial intelligence (AI) and its mathematical models have also been applied to new157 housing price valuation and prediction models. Machine learning, a branch of AI, has been158 under development since the 1950s. In 1980, a breakthrough occurred with the Hopfield159 model, the first neural network created by John Hopfield, whose neurons were all input160 and output, fully connected to each other in both directions [17].161 In the 1980s, Hinton and Sejnowski developed machines that combined Hopfield162 neural networks with the simulated annealing algorithm, resulting in a two-layer fully163 connected neural network [18]. Around 1995, Vapnik and Cortés achieved one of machine164 learning’s most significant advancements: Support Vector Machines (SVM). These ma-165 chines map input vectors non-linearly into a high-dimensional feature space, enabling the166 construction of a linear decision boundary and yielding accurate results [19].167 At the beginning of the 21st century, Breiman (2001) introduced another machine learning168 method, Random Forest, which combines several decision trees. Each tree is created from a169 random subset of features, with nodes generated using randomly selected characteristics [20].170 Machine learning models have advanced further, exploring new directions. Com-171 bining multiple models has led to new methodologies widely used by researchers across172 various fields, including housing price prediction. In these models, computers identify173 patterns in data without being explicitly programmed for the task, mimicking human174 reasoning [9]. The most commonly used models include linear and logistic regression,175 Machine learning models have advanced further, exploring new directions.176 Combining multiple models has led to new methodologies widely used by researchers177Urban Sci. 2025, 9, 32 across various fields, including housing price prediction. In these models, computers 4 of 20178 identify patterns in data without being explicitly programmed for the task, mimicking179 human reasoning [9]. The most commonly used models include linear and logistic180 decision trees,181 regression, Random182 decision Forest,183 trees, Bayesian184 Random Forest,networks,185 Bayesian AdaBoost, Support Vector186 networks, AdaBoost, Machines,187 Support Vector188 XGBoost, K-Nearest Neighbors, and Artificial Neural Networks [21].189 Machines, XGBoost, K-Nearest Neighbors, and Artificial Neural Networks [21].190 In Figure191 In Figure1,1,a aclassification of of192 classification thethe193 different advanced194 different approaches195 advanced for valuing196 approaches real estate197 for valuing real198 prices is presented: HPM, neural networks, spatial analysis, and so on.199 estate prices is presented: HPM, neural networks, spatial analysis, and so on.200201202 ADVANCED REAL ESTATE203 VALUATION METHODS204205206207208 HEDONIC PRICES SPATIAL ANALYSIS MULTICRITERIA THEORY209 MACHINE LEARNING210 (regression models)211212213214215 Supervised Learning Unsupervised Learning Reinforcement Learning216 (predictive models) (descriptive models) (based on findings, maximizes them)217218219220221 Most used models:222 Most used models: K-Means Clustering223 Neural Networks Probabilistic Clustering Methods224 Regressions (linear/multiple)225 Decision Trees226 Random Forest227 XGBoost228 Support Vector Machines229 ARMA and ARIMA models230 K-Nearest Neighbors231 Bayesian Networks232233234235 Figure236 Figure 1.237 1. Most238 Most commonly239 commonly used240 used advanced prediction models241 advanced prediction models (source:242 (source: own243 own elaboration).244 elaboration).245246 valuationmethods247 Advanced valuation methods in real248 in the the real249 estate estate sector250 sector are are into251 classifiedclassified252 four into four253 categories:254 categories:255 hedonic hedonic256 price modelsprice modelsmodels257 (regression (regression models the258 that estimate thatvalue259 estimate the value260 of each of each261 characteristic of262 characteristic263 a property), of a264 machineproperty),265 learningmachine266 models learning267 (based onmodels (based on techniques268 mathematical mathematical techniques269 and artificial270 intelligence),271 and artificial spatial analysisspatial272 intelligence), modelsanalysis273 (which focus274 models on (which275 estimating the on276 focus characteristics277 estimating of thea278 specific area), of279 characteristics anda specific280 multi-criteria theory models281 area), and (which282 multi-criteria estimate283 theory modelsall types284 (whichofestimate285 intangible286 all287 property288 types characteristics).289 of intangible property characteristics).290 Machine learning is classified into three types of learning: supervised supervised learning, which291 consists of292 of predictive293 predictivemodels294 modelsand andforms295 formsthe296 the basis of of297 basis thethe298 analysis in in299 analysis thisthis300 review (neural301 review net-302 (neural303 works, decision304 networks, trees, trees,305 decision XGBoost, etc.);306 XGBoost, unsupervised307 etc.); learning, which308 unsupervised focuses309 learning, on analyzing310 which focuses andon311 predicting descriptive312 analyzing and models; and313 predicting reinforcement314 descriptive models; and learning, which aims to315 reinforcement maximize316 learning, findings.317 which aims to318 Thanks to319 maximize their problem-solving capabilities, the models that make up these methods320 findings.321 are efficient alternatives for322 Thanks to their housing price323 problem-solving prediction,324 capabilities, the modelsas they325 that can identify326 make up thesenon-linear327 methods328 relationships329 are efficient between330 alternativesrelevant331 for data.332 housing Theprice333 development334 prediction,of as335 these models336 they can helps predict337 identify future338 non-linear339 housing prices and340 relationships betweenestablish real data.341 relevant estateThepolicies [10].342 development of these models helps predict343 future housing prices and establish real estate policies [10].344 3. Methodology for the Systematic Review of Housing Price345 Estimation Models346 A systematic review integrates various studies that analyze the same question, consid-347 ering these studies in an observational manner to assess the trajectory of the analysis on348 3. Methodology for the Systematic Review of Housing Price Estimation349Urban Sci. 2025, 9, 32 Models 5 of 20350 A systematic review integrates various studies that analyze the same question,351 considering these studies in an observational manner to assess the trajectory of the analysis on352 the topic353 the topic being354 being reviewed.355 reviewed. TheThe main356 main objective357 objective of of aa systematic358 systematic review359 review is is to360 to gather361 gather thethe highest362 highest363 number of original studies on the topic being analyzed, in order to provide a high level364 number of original studies on the topic being analyzed, in order to provide a high level of365 of366 scientific evidence367 scientific evidence [22].[22].368 Once the369 Once therelevant370 relevantstudies371 studiesforfor372 thethe research373 research areare located,374 located, theythey are analyzed375 are analyzed comprehen-376 comprehensively377 sively378 to to compare379 compare them and them and identify380 identify their mosttheirimportant381 most important382 aspects aspects [23]. Systematic383 [23]. Systematic reviews reviews384 provide385 provide a fusion of information on a specific topic, which is essential386 a fusion of information on specific topic, which is essential for future research on for future research387 the same on388 the same389 subject andsubject andanswers390 can offer can offertoanswers391 questionstothatquestions392 could not thatbecould not through393 obtained be obtained through394 independent395 independent studies. Thus, systematic reviews are highly396 studies. Thus, systematic reviews are highly useful to researchers when addressinguseful to researchers when397 the398 addressing the situation of a topic they wish to399 situation of a topic they wish to study and analyze [24,25]. study and analyze [24,25].400 Systematic reviews401 Systematic reviews areare often402 often conducted403 conducted with with the404 the help405 help of of the406 the PRISMA407 PRISMA Statement.408 Statement. It It is409 is410 essentially a guide to assist authors in conducting this type411 essentially a guide to assist authors in conducting this type of study [26]. of study [26].412 This study413 This study used414 used aa systematic415 systematic review416 review to to examine417 examine the the number418 number of of existing419 existing articles420 articles that421 that422 present practical cases of various advanced prediction models, specifically423 present practical cases of various advanced prediction models, specifically in the real estate in the real estate424 sector, in425 sector, in aa comparative426 comparative manner,427 manner, seeking428 seeking to to identify which of429 identify which of these430 these models431 models offers432 offers the433 the best434 best435 results for predicting housing prices (see436 results for predicting housing prices (see Figure 2). Figure 2).437438439440441 Figure442 Figure 2.443 2. Flowchart444 Flowchart of445 of the article selection446 the article selection process447 process for448 for systematic449 systematic review—PRISMA450 review—PRISMA (source:451 (source: own452 own453 elaboration).454455 The articles used for this review were retrieved electronically from the Scopus, Google456 Scholar, and Web of Science platforms in May 2024; more specifically, the search was457 conducted in Scopus from 1 May to 10 May, in Google Scholar from 10 May to 20 May, and458 Urban Sci. 2025, 9, 32 6 of 20459460461462 in Web of Science from 20 May to 30 May. These three platforms are important databases463 for retrieving studies, as they hold a wide range of documents categorized across different464 scientific fields.465 Several combinations of keywords in English were entered into these three databases466 to locate as many studies on the topic as possible for analysis in the review. The terms used,467 among others, for the search were “housing price prediction”; “housing price models”;468 “machine learning”; “housing prices”; “supervised learning”, “hedonic prices models”,469 “artificial neural networks”, “support vector machine”, and “random forest”.470471 3.1. Search Criteria472 The exploration of the literature on the topic of interest was carried out longitudinally473 over time, searching for studies that have been published since the early 21st century,474 including preprints. The search was conducted so that the keywords in the documents475 would appear in the title, abstract, or in the keywords of the articles. Similarly, only476 documents in English or Spanish were selected. Only articles were chosen, excluding book477 chapters, technical reports, dissertations, editorials, opinions, and conference proceedings.478479 3.2. Eligibility Criteria480 The articles included in the review meet the following requirements:481 • They conduct housing price prediction studies using more than one advanced predic-482 tive model.483 • They have been published in the 21st century, i.e., from the year 2000 to 2024, to track484 advancements in the use of these techniques over time.485 • They use empirical data, meaning the models have been applied practically and are486 not just explained theoretically.487 • They provide a comparison between these models.488 • The data used must be obtained from reliable sources, whether public organizations or489 companies dedicated to the real estate sector, to provide empirical housing price data.490 Articles that use data obtained through the “web scraping” technique are excluded, as491 this method provides potential sale price data rather than the actual transaction value of492 the property.493494 4. Study Selection and Results495 The search for documents on housing price prediction models yielded a total of 516496 studies. After conducting the necessary screening, excluding duplicate articles and those497 that did not meet the inclusion criteria, a total of 23 articles were selected for inclusion498 in this systematic review. The selection of articles has been supervised by two reviewers499 throughout the procedure, including a third reviewer in the final phase to resolve any500 disagreements that the first two might have.501 Figure 2 shows the process of selecting the articles included in this review and high-502 lights those that were excluded. After removing non-article studies and duplicate articles,503 further exclusions were made for various reasons. First, 212 articles were eliminated for504 lacking essential requirements for the analysis, such as failure to identify the source of the505 data or the year in which the study was conducted, among others. The remaining records506 were then subjected to eligibility criteria, resulting in the exclusion of 43 articles for being507 conducted using the “Web Scraping” technique; 52 articles were excluded because they508 focused on prediction models for fields other than real estate, such as mobile telephony or509 tourism, among others; 21 articles were excluded for not comparing two or more prediction510 models; and finally, 28 articles were removed for presenting only the theoretical part of the511 Urban Sci. 2025, 9, 32 7 of 20512513514515 models without conducting an experimental analysis. In this way, the 23 articles included516 in this systematic review were obtained.517518 4.1. General Characteristics of the Included Studies519 Table 1 outlines the main characteristics of the articles included in this systematic520 review. In terms of the number of authors, it is notable that 31% of the selected articles521 were co-authored by three authors, and 26% by four authors. Despite the challenges522 involved in conducting predictive price analysis for housing—due to the complexity of523 obtaining, tabulating, and adapting data, as well as executing predictive models in com-524 puter applications—it is worth highlighting that 22% of the articles were written by a single525 author. A smaller percentage, 17%, had two co-authors, and only one article was authored526 by six people, representing 4% of the total articles included in the review.527528 Table 1. Characteristics of the articles selected for the review.529530 Number of Articles Included 23531 Authors532 1 author 5533 2 authors 4534 3 authors 7535 4 authors 6536 5 authors 0537 6 authors 1538 Country539 Kosovo 1540 South Korea 2541 Australia 2542 Poland 1543 USA 3544 Taiwan 4545 United Kingdom 1546 Turkey 2547 China 2548 Slovenia 1549 Thailand 2550 Spain 2551 Year from publication552 From 2000 to 2010 1553 From 2011 to 2020 8554 From 2021 to 2024 14555 Type of study556 Cross-sectional 8557 Longitudinal 15558 Data Sample Analysis Period559 From 1 month to 5 years 17560 From 6 years onwards 6561 Sample Size562 From 0 to 5000 9563 From 5000 to 10,000 3564 From 10,000 onwards 11565 Urban Sci. 2025, 9, 32 8 of 20566567568569 Table 1. Cont.570571 Number of Articles Included 23572 Number of models compared in the study573 Two models 7574 Three models 7575 Four models 4576 Five models 2577 Seven models 1578 Eight models 1579 Eleven models 1580 Source: own elaboration.581582583 Most empirical applications have been carried out in Asia, comprising 43% (10 studies),584 with Taiwan being the country that has published the most research on advanced housing585 price prediction models. A total of 26% of the articles were from European countries,586 13% from America, specifically the United States, and 9% from both Australia and Turkey587 (Middle East), which are the geographical regions with the fewest articles. As observed,588 the Asian continent is a pioneer in this type of methodology in the housing market due589 to its significant technological advancements and the ease of obtaining macro data from590 official organizations [27].591 The study design was predominantly longitudinal (15 out of 23 articles), while only592 eight were cross-sectional. Regarding the time periods of the data used in the analysis, 74%593 of the articles collected data over periods ranging from one month to five years, while 26%594 focused on longer periods, starting from six years onward.595 Concerning the comparative study analysis between advanced prediction models,596 seven articles compared the results of using two models, seven others compared three597 models, four articles examined four models, and two articles compared five models. Ad-598 ditionally, two articles compared seven and eight models, respectively, while one article599 compared up to eleven advanced prediction models.600 Finally, it is worth noting that most of the publications were made from 2020 onwards,601 with 61% of the articles published during this period. This could be attributed to advance-602 ments in technologies and mathematical techniques in recent years, which have led to603 a surge in comparative studies of price prediction models, aiming to better understand604 how both internal and external variables influence housing price fluctuations. Only one605 article from the early 2000s was included in this systematic review, with the rest published606 between 2011 and 2020.607608 4.2. Advanced Housing Price Prediction Models609 In the previous section, it is noted that a large number of comparative models have610 been used in the various articles, with some including up to seven or eight prediction611 models. In fact, one study compares up to eleven models, primarily due to the number of612 observations involved in the analysis of each method. On the other hand, many articles613 compare only two or three models.614 In total, twenty-three different prediction models have been used to carry out varia-615 tions in housing prices based on both internal and external characteristics, including, in616 some studies, variables related to the neighborhood, location, availability of surrounding617 services, and environmental characteristics. Technological advancements have enabled the618 use of applications that enhance the prediction of these models, given the large number619 of observations they include, allowing for research that is highly useful for entrepreneurs,620 investors, and financial institutions interested in understanding fluctuations in real es-621 Urban Sci. 2025, 9, x FOR PEER REVIEW 9 of 18622 Urban Sci. 2025, 9, 32 9 of 20623624625626 fluctuations627 tate prices in based628 real estate prices based629 on specific on specific characteristics.630 characteristics. In fact, many In of fact,631 thesemany of these areand632 are variations633 variations and of the634 combinations combinationsmost of the most635 widely usedwidely636 models,used models,637 as shown inas shown638 Table 1. in Table 1.639 Figure 3 illustrates the types of models used in the studies640 Figure 3 illustrates the types of models used in the studies of this of this review, as well641 review, as as642 as well643 howhow often these644 often models645 these modelsappeared646 appearedin the various647 in the articles.648 various TheThe649 articles. most frequently650 most usedused651 frequently model,652 model,653 appearing in fourteen different articles, is the Random Forest model. This654 appearing in fourteen different articles, is the Random Forest model. This model consists of model consists655 of the656 thecreation657 creation and and combination658 combination of several decision659 decision trees,660 trees, with661 with the662 the aim663 aimof ofcombining664 combiningthe theout-665 outputs of these trees into a single single output666 output toto address667 addressregression668 regressionor orclassification669 classificationproblems670 problems[28].671 [28].InInfact,672 fact,ininfive673 fiveofofthe674 thethirteen675 thirteenarticles676 articleswhere677 whereititappears,678 appears,this679 thismodel680 modelwaswaschosen681 chosenasasthe682 theone683 onethat684 thatprovides685 providesthe themost686 most significant results compared to the others, due to statistical687 significant results compared to the others, due to statistical results688 results689 suchsuchas Ras 2 orR2the690 or the absolute691 absolute error.692 error.693694 More than 10 times •The only model that has been used more than 10 times,695 n=1 specifically 14 times, is the Random Forest model.696697698699 •Linear Regeession700 Between 5 and 10701 times •Decision Tree702 •Hedonic Price Model703 n=4704 •Artificial Neural Networks705706707708 •Support vector regression709 •Gradient-Boosted tree710 •Lasso regression711 •Ridge regression712 Less than 5 times •Regression trees713 n = 10 •Gradient boosting machine714 •Support Vector machine715 •LightGBM716 •XGBoost717 •Elastic net718719 •K-Nearest neighbors720 •Spatial analysis721 •Genetic Algorithm (GA)722 1 time •Gaussian regression723 •Cubist model724 n=8725 •Hybrid model726 •Gradient Boosting727 •AdaBoost728729730 Figure 3. Number of times prediction models appear in the articles (source: own elaboration).731 Figure 3. Number of times prediction models appear in the articles (source: own elaboration).732 The models that appear most frequently (between five and ten times) in the different733 The models734 articles are thethat735 mostappear most used736 commonly frequently (between737 in prediction five and738 studies. teninclude739 These times) in the different740 regression models,741 articles are the742 decision trees,most commonly743 HPMs, and used in744 artificial neuralprediction745 networks.studies. These include regression746 models, Next,747 decision trees,748 there are HPMs, and that749 ten models artificial750 appearneural networks.751 in fewer than five articles, either because they752 Next,753 are lessthere are ten or754 well-known models that their755 because appear in fewer than756 predictions fiveas757 are not articles,758 accurateeither759 or because they760 significant as the761 are previous762 less well-known or because their predictions are not as accurate or significant763 ones. These include XGBoost, regression trees, support vector machines, and as the764 previous765 Lasso ones. These among766 regression, includeothers.767 XGBoost, regression768 Finally, seven trees, support769 models are thevector machines,770 least frequent andandleast771 Lasso regression, among others. Finally, seven models are the least frequent772 known in the field of housing price prediction, as they appear in only one article. and least773 known in the field of housing price prediction, as they appear in only one article.774 4.3. Most Influential Variables in Housing Price Prediction Models775 To use advanced housing price prediction models, it is necessary to include not only a776 sample of housing transactions but also a significant number of variables (both internal777 Urban Sci. 2025, 9, 32 4.3. Most Influential Variables in Housing Price Prediction Models 10 of 20778 To use advanced housing price prediction models, it is necessary to include not only779 a sample of housing transactions but also a significant number of variables (both internal780 and781 and external)782 external) related783 related toto the784 the properties785 properties in786 in order787 order to788 to understand789 understand howhow prices790 prices vary791 vary based792 based on793 on794 these795 these aspects. Figure 4 shows the variables from the articles in the systematic review that796 aspects. Figure 4 shows the variables from the articles in the systematic review that797 have798 have proven799 provento to be800 be the801 the most802 most significant803 significant and,804 and, therefore,805 therefore,have806 havehad807 had the808 the greatest809 greatestimpact810 impacton811 on the812 the813 process of housing price variation and prediction.814 process of housing price variation and prediction.815816817 Most significant variables818 FINANCIAL POLICIES 2.4%819 DISTANCE TO WORK 2.4%820 DISTANCE TO MAJOR ROADS 4.8%821 ENVIRONMENTAL FACTORS 9.5%822 NEIGHBORHOOD 11.9%823 DISTANCE TO TRANSPORTATION… 14.3%824 PROPERTY CHACARTERISTICS 16.7%825 DISTANCE TO THE CENTER 16.7%826 LOCATION OF THE PROPERTY 21.4%827828 0.0% 5.0% 10.0% 15.0% 20.0% 25.0%829830831 Figure4.832 Figure 4. Most833 Most influential834 influential variables835 variables in836 in housing837 housing price838 price variation839 variation in840 in the841 the articles842 articles of843 of the844 the systematic845 systematic review846 review847 (source: own elaboration).848 (source: own elaboration).849850 According851 According to to Figure852 Figure 4,4, the853 the variables854 variables that that have855 have proven856 proven to to be857 be the858 the most859 most significant860 significant and and861 influential on housing prices, and those that make properties862 influential on housing prices, and those that make properties more attractive to buyers more attractive to buyers863 according864 according to to the865 the models866 models analyzed867 analyzed in in the868 the selected869 selected articles,870 articles, are871 arenine.872 nine. TheThe variable873 variable that that has874 has875 been repeated the most (up to nine times) and, therefore,876 been repeated the most (up to nine times) and, therefore, can be considered the most can be considered the most877 influential,878 influential,isisthe the location879 locationof of the880 the property,881 property,which whichincludes882 includesthe thearea/neighborhood883 area/neighborhood in in which884 which885 itit is located, as well as the characteristics886 is located, as well as the characteristics of that area. of that area.887 Next,888 Next, following889 following the the order890 order of of importance,891 importance, there there are892 aretwo893 two variables894 variables thatthat have895 have appeared896 appeared897 seven times each: distance to the central commercial district,898 seven times each: distance to the central commercial district, which includes the distance which includes the distance to899 the city center and the services it offers, such as shopping centers,900 to the city center and the services it offers, such as shopping centers, hospitals, educational hospitals, educational901 institutions,902 institutions,etc.; etc.;structural903 structuralcharacteristics904 characteristics ofof thethe905 property,906 property, which907 whichinvolve908 involvefactors suchsuch909 factors as theas910 size, number of rooms, presence of an elevator, outdoor911 the size, number of rooms, presence of an elevator, outdoor spaces, and others. spaces, and others.912 The913 The variable914 variable considering915 considering distance916 distance to to transportation917 transportation infrastructure,918 infrastructure, i.e.,i.e., the919 the distance920 distance921 to922 to train stations, trams, buses, etc., has been one of the most significant, appearing in923 train stations, trams, buses, etc., has been one of the most significant, appearing in six924 six925 studies. In five of the articles, the neighborhood and its characteristics926 studies. In five of the articles, the neighborhood and its characteristics were considered were considered927 one928 one of of the929 the most930 most influential931 influential variables932 variables in in the933 themodels.934 models. ThisThis variable935 variable includes936 includes economic,937 economic,938 ethnic/racial conditions,939 ethnic/racial conditions, etc. etc.940 Environmental941 Environmentalfactors factorsofofthe942 thelocation,943 location,which whichinclude944 includegreen945 greenspaces,946 spaces,gasgas emissions,947 emissions,and and948 noise levels, among other aspects, in the geographical area949 noise levels, among other aspects, in the geographical area where the property is located, where the property is located,950 have951 havebeen beensignificant952 significantininfour fourarticles.953 articles.AAfactorfactor that954 thathas955 hasbeen956 been repeatedly957 repeatedly mentioned958 mentioned twice as959 twice960 influential961 as influential is theisdistance from the962 the distance from property to majortoroads.963 the property majorLastly,964 roads.with only with965 Lastly, one incidence,966 only one967 two968 incidence, two variables appear: distance from the property to workplaces policies,969 variables appear: distance from the property to workplaces and financial and financialwith970 the latter with971 policies, including the conditions972 the latter includingset thebyconditions973 banks when setapplying974 by banksforwhen a mortgage975 applying loanforfora976 property acquisition.977 mortgage loan for property acquisition.978979 5. Discussion980 In order to present all the studies selected for this systematic review, Appendix A981 provides a concise, detailed overview of these studies, including authors, the period in982 5. Discussion983 In order to present all the studies selected for this systematic review, Appen984Urban Sci. 2025, 9, 32 11 of 20985 provides a concise, detailed overview of these studies, including authors, the per986 which the analysis took place, sample sizes, compared prediction models, and a987 summary/conclusions of the objective of each study.988 which the analysis took place, sample sizes, compared prediction models, and a brief989 In the selected articles for this systematic review, the various authors choose990 summary/conclusions of the objective of each study.991 models whose results provide the best fit for housing price prediction based o992 In the selected articles for this systematic review, the various authors choose those993 corresponding parameters. They focus on statistical measures that demonstrate th994 models whose results provide the best fit for housing price prediction based on the corre-995 namely selecting996 sponding parameters. They focustheonmodel that achieves997 statistical measuresboth that ademonstrate998 higher R2, which999 this fit,measures1000 namely how we1001 model fits the data, as well as the lowest1002 2 levels of Mean1003 selecting the model that achieves both a higher R , which measures how well the modelAbsolute Error (MAE) and1004 fits the data, asMean1005 well Square Error levels1006 as the lowest (RMSE). Integrating1007 of Mean Absolute bothError1008 MAE(MAE)1009 and RMSE1010 and RootintoMean1011 prediction m1012 helps to better understand the ability of different models1013 Square Error (RMSE). Integrating both MAE and RMSE into prediction models helps to to predict prices more accu1014 [29].the ability of different models to predict prices more accurately [29].1015 better understand1016 In Figure 5, theIndifferent1017 Figure 5,particularities,1018 the different bothparticularities,1019 positive and both positive1020 negative, of and negative,1021 the main ad- of the1022 advanced models1023 vanced price prediction price prediction models1024 in the housing in the1025 market are housing market1026 presented, are1027 such as thepresented,1028 hedonic such a1029 price models and machine learning models, as discussed throughout this article. These this a1030 hedonic price models and machine learning models, as discussed throughout1031 characteristicsThese characteristics1032 are analyzed throughare analyzed1033 a SWOT through1034 analysis, a SWOT analysis,1035 commonly commonly1036 used to identify weak- used to id1037 weaknesses, threats, strengths, and opportunities. This1038 nesses, threats, strengths, and opportunities. This will help to understand the positivewill help to understan1039 positiveofand1040 and negative aspects the negative aspectsprediction1041 main advanced of the main advanced1042 methods andprediction1043 allow for amethods1044 thorough and allow1045 thorough1046 comparison of them. comparison of them.10471048104910501051 Figure 5. Characteristics of HPMs and machine learning models through the SWOT analysis (source:1052 own elaboration).1053 Figure 5. Characteristics of HPMs and machine learning models through the SWOT analysis (s1054 ownof1055 In the works elaboration).1056 [2,30–36], where all authors focus on the analysis of HPM, they agree1057 that, when compared to machine learning models, the latter offer better adjustments in1058 prediction models. According to these authors, this is due to the common occurrence of1059 multicollinearity in HPM and the generation of indices that this model requires, which1060 leads to a loss of information when interpreting results. In some cases, HPM can also face1061 heteroscedasticity problems, though these can be corrected, and the variables may still be1062 significant. HPM helps quantify both the internal and external characteristics of a property1063 Urban Sci. 2025, 9, 32 12 of 201064106510661067 through regression techniques; however, this model has limitations, such as the inclusion of1068 many explanatory variables, which leads to the aforementioned multicollinearity problems,1069 as well as the inability to find non-linear relationships between different variables [37].1070 However, these limitations can be corrected through the use of machine learning1071 algorithms, which are more flexible, do not have specific requirements regarding data1072 distribution, and can find non-linear relationships. For this reason, many authors use such1073 methods to resolve the previously mentioned issues of HPM [38]. Thus, it is concluded1074 that machine learning methods are a good complement to HPM, and by combining both1075 techniques, the results can achieve a higher degree of fit. In fact, in [12], HPMs are seen as1076 a solid model for highlighting the characteristics that most influence housing prices, and1077 these authors aim to address the non-linear relationships in hedonic prices by performing a1078 robustness test with a machine learning model. Similarly, [39] compare HPMs with spatial1079 analysis, finding that the latter model is a good complement to hedonic prices, as both1080 models provide results with a higher level of significance than each model individually.1081 In summary, in these works, the hedonic methodology is fundamental for understand-1082 ing the variation in housing prices based on the bundle of characteristics it possesses. If the1083 characteristics of a property are known and its prices can be estimated, HPM proves to be1084 very useful and provides an opportunity to estimate the value [40].1085 Regarding models based on tree algorithms, such as decision trees, random forest, and1086 gradient-boosted trees, these are the models most frequently selected as the most significant1087 in the articles of this review. These models offer advantages in terms of performance with1088 large datasets [41]. In his article, [42] confirms that the decision tree model is the one that1089 best fits housing price prediction, both in terms of achieving the highest R2 and the lowest1090 error level. This study suggests that, most of the time, more complex models are given1091 more consideration, but simpler models should also be observed more closely, as they can1092 yield very similar results.1093 The random forest model is the one that best fits price prediction in most articles that1094 use it, compared to other models. In studies like those of [31,43,44], the random forest1095 model showed very significant results for housing price prediction and performed better1096 than the other models, both in training and test data.1097 However, in the studies of [32,35,36], when comparing random forest with HPM and1098 finding the former with better results, the authors confirm that random forest adequately1099 reflects non-linearity and the real complexity of real estate markets and that, therefore, the1100 random forest model is a good complement to HPM. Similarly, studies like [45] justify that1101 the best solution is an ensemble classifier model, i.e., the combination of decision tree with1102 random forest and gradient-boosted tree, as it shows excellent results.1103 Other studies confirm the good performance of tree-based models, such as those used1104 to predict the housing price index in the US, where the application of a random forest model1105 achieved an error margin of only 5% [46]. Or studies showing the effectiveness of random1106 forest compared to multiple regression models in the housing market in Australia [47].1107 Regarding neural networks, these stand out in studies comparing them with HPM, as1108 evidenced in the works of [2,27,30,34]. As mentioned earlier, neural networks provide more1109 precise results than HPM in predicting housing prices. Moreover, in studies predicting1110 housing prices in large cities, neural networks have been proven effective for this task, such1111 as in the case of Taranto, Italy [48]. However, there have also been studies where neural1112 networks were not effective, such as in the prediction of housing prices in Iran, where the1113 fuzzy regression model proved more effective than neural networks [49].1114 The other models chosen by authors as the best models in their studies, such as1115 XGBoost or LightGBM, were selected for providing better results in parameters and being1116 more adaptable to the model.1117 Urban Sci. 2025, 9, 32 13 of 201118111911201121 Similarly, within the housing market and urban planning, artificial neural network1122 models, random forests, and support vector machines are frequently used in land use1123 classification aspects and are even employed to simulate land use planning processes [50].1124 It is important to note that many authors choose the models to develop based on1125 the type of variables they want to include in the model, considering them significant for1126 housing price prediction. Most of them use variables related to the external characteristics1127 of the property, especially the location, such as neighborhood characteristics, environmental1128 context, or the perception of the area and neighborhood held by potential buyers.11291130 6. Conclusions1131 Access to housing is closely linked to its affordability, which is determined by the1132 characteristics and environment of its location. The determination and understanding of1133 the prices surrounding a property are fundamental aspects when it comes to accessing1134 housing. This is why this systematic review was conducted, with the goal of gathering1135 information on the set of internal and external characteristics that contribute to determining1136 the price of a property.1137 After conducting the systematic review to understand the extent of studies focused1138 on price prediction in the real estate sector, it has been found that advanced and machine1139 learning models are frequently used in this sector. The importance of hedonic regressions1140 in property valuation is emphasized, as this type of model contributes significantly to1141 understanding how internal and external variables of a property affect the overall price1142 and how it varies depending on each of these factors. It is also concluded that combining1143 hedonic price models (HPM) with machine learning models results in more meaningful1144 outcomes for property price prediction. Therefore, it is advisable to merge this advanced1145 method with a machine learning model.1146 Machine learning-based prediction models, inherent to artificial intelligence (AI),1147 provide an objective viewpoint and can offer strong results in predicting real estate prices.1148 The use of these models, particularly considering external variables of properties (such1149 as proximity to points of interest, neighborhood, environmental context, etc.), can deliver1150 results that closely reflect real market prices with minimal margin of error. As such,1151 these techniques can be a useful complement for real estate professionals, as well as for1152 governments or public institutions in developing economic or housing policies.1153 The use of predictive algorithms with macro data can analyze variables that were1154 previously not measured, providing clarity when establishing public policies, although1155 they must be used with great caution. By combining new technologies with human capital,1156 research in the real estate market field would be more fluid and yield better results in large1157 cities. In this way, all market participants could understand the value of each property1158 attribute and estimate the price fluctuations of each attribute separately, as well as the1159 overall value of the property. This would help avoid the implementation of prices that1160 diverge from the market reality, which are set using traditional valuation and prediction1161 methods, and thus, have a high degree of subjectivity.1162 By analyzing the various studies on this topic, it has been found that much more1163 research has been conducted on these prediction models in the real estate sector during the1164 second decade of the 21st century. This leads to the conclusion that in the coming years,1165 a wide variety of studies will be carried out with different samples, time periods, and1166 advanced prediction models in this field. These studies will introduce new ideas and add1167 data to help make this market more transparent, balanced, and accessible to everyone.11681169 Author Contributions: Conceptualization, I.M.-F.; M.-T.S.-M. and M.P.-E.; methodology, I.M.-F.;1170 M.-T.S.-M. and M.P.-E.; investigation, I.M.-F.; M.-T.S.-M. and M.P.-E.; writing—original draft prepa-1171 ration, I.M.-F.; M.-T.S.-M. and M.P.-E.; writing—review and editing, I.M.-F.; M.-T.S.-M. and M.P.-E.;1172 Urban Sci. 2025, 9, 32 14 of 201173117411751176 visualization, I.M.-F.; supervision, I.M.-F. All authors have read and agreed to the published version1177 of the manuscript.11781179 Funding: This research received no external funding.11801181 Data Availability Statement: The original contributions presented in this study are included in the1182 article. Further inquiries can be directed to the corresponding author.11831184 Conflicts of Interest: The authors declare no conflicts of interest.11851186 Appendix A. Description of the Selected Studies1187 Table A1. Comparative studies of advanced price prediction models.11881189 Authors Compared Prediction Models Summary/Conclusions1190 The prediction of housing prices in the1191 city of Prishtina is analyzed through a1192 dataset that includes housing prices and1193 their characteristics, by comparing four1194 Hoxha, 2024 [42] Linear Regression, Decision Tree,1195 machine learning models to extend the1196 Analysis period: 2019–2023 K-Nearest Neighbors, and Support1197 resulting knowledge to other real estate1198 Sample size: 1512 Vector Regression.1199 markets. The model that provides the best1200 ability to predict prices is the Decision1201 Tree model, as it achieves the lowest MSE1202 and the highest R2 .1203 They analyze how housing prices in the1204 metropolitan area of Adelaide vary based1205 on the impact of certain characteristics, in1206 this case, adding a spatial variable. It is1207 demonstrated that non-linear models1208 Soltani et al. 2022 [47] Linear Regression, Decision Tree,1209 perform better than linear ones, and that1210 Analysis period: 1984–2016 Random Forest, and1211 applying a spatio-temporal variable is1212 Sample size: 428,000 Gradient-Boosted Tree.1213 effective in machine learning models. The1214 model with the best performance is the1215 Gradient Boosted Tree, achieving the1216 highest R2 and the lowest RMSE1217 and MAE.1218 Understanding the preferences of1219 homebuyers based on the environmental1220 characteristics of the properties. By1221 Ligus and Peternek, 2017 [39] applying two econometric models, it is1222 Analysis period: 2012–2014 HPM and spatial analysis. found that geographic characteristics are1223 Sample size: 6318 significant and cause variation in prices.1224 The union of both models together1225 presents more significant results than the1226 HPM alone.1227 It aims to address the need to include1228 uncommon variables in housing price1229 variation, such as the effects and1230 characteristics of neighborhoods, as well1231 Chen et al. 2023 [51] Gradient-Boosted Tree, Random1232 as the economic and ethnic conditions1233 Analysis period: 2014–2017 Forest, Elastic Net, Lasso Regression,1234 within them. As a result, transportation1235 Sample size: 25,135 and Ridge Regression.1236 and the economic and sociodemographic1237 characteristics of the neighborhood are1238 significant, but the economic ones stand1239 out as the most significant.1240 Urban Sci. 2025, 9, 32 15 of 201241124212431244 Table A1. Cont.12451246 Authors Compared Prediction Models Summary/Conclusions1247 By applying these three models, it aims to1248 address the problems of housing price1249 prediction. Ensemble regression trees1250 Lahmiri et al. 2023 [52] Ensemble Regression Trees, Support achieve the best results for prediction as1251 Analysis period: 2012–2013 Vector Regression, and they exhibit the smallest error rate and1252 Sample size: 414 Gaussian Regression. low error variability, indicated by the1253 range of the distribution. This makes them1254 strong candidates for future predictions1255 in Taiwan.1256 The Google Street View tool is used to1257 analyze the environmental location factors1258 in housing prices through a neural1259 HPM, Neural Network, Lasso1260 Wang, 2023 [2] network model. As a result, this model1261 Regression, Ridge Regression,1262 Analysis period: 2015–2020 improves accuracy and provides more1263 Regression Trees, Random Forest,1264 Sample size: 34,447 significant results than the others, this is1265 and Gradient Boosting Machine.1266 because it recognizes edges and more1267 complex shapes in its layers and1268 highlights important features.1269 Using images of points of interest and1270 neighborhood characteristics, this study1271 examines how housing prices vary based1272 on the image of the property compared to1273 Chen et al. 2022 [31]1274 HPM, Random Forest, and Gradient its intrinsic features. In general, the R21275 Analysis period: 2013–20191276 Boosting Machine. improves across all models when images1277 Sample size: 137,1321278 are introduced, but the random forest1279 model exhibits the smallest error rate,1280 making it the one that delivers the most1281 signif-icant results.1282 It investigates the Random Forest model1283 as a predictor of housing prices and1284 compares it with HPM to assess the1285 Hong et al. 2020 [32] results. Between the two models, Random1286 Analysis period: 2006–2017 HPM and random forest. Forest proves to be more adaptable to the1287 Sample size: 16,601 reality of the real estate market, as it1288 achieves a higher R2 and a lower MAPE,1289 random forest is proposed as an excellent1290 complement to the HPM.1291 It uses the Support Vector Machine1292 method to predict housing prices,1293 incorporating housing variables in the1294 Chen et al. 2017 [33]1295 hedonic model to verify price variations.1296 Analysis period: 2007–2010 HPM and Support Vector Machine.1297 The support vector machine model1298 Sample size: 3.9911299 demonstrates strong predictive power in1300 forecasting housing prices with1301 high accuracy.1302 Urban Sci. 2025, 9, 32 16 of 201303130413051306 Table A1. Cont.13071308 Authors Compared Prediction Models Summary/Conclusions1309 It analyzes the determinants of housing1310 prices, including location. It compares1311 both methods and finds that neural1312 Selim, 2009 [34]1313 HPM and Artificial networks are the better alternative for1314 Analysis period: 20041315 Neural Networks. price prediction as they are more accurate,1316 Sample size: 5.7411317 although the HPM heteroscedasticity1318 correction results in most variables being1319 highly significant.1320 It uses these three models for housing1321 price evaluation. After applying the1322 models, random forest and gradient1323 Ho et al. 2020 [43] Random Forest, Support Vector boosting machine performed better than1324 Analysis period: 1996–2004 Machine, and Gradient support vector machine (higher R2 and1325 Sample size: 39,554 Boosting Machine. lower error rate). However, the latter is1326 still considered a very useful algorithm,1327 providing accurate predictions within a1328 strict time constraint.1329 It analyzes housing price prediction using1330 three models, both advanced and machine1331 Begum et al. 2022 [44] learning-based. After analyzing the three1332 Linear regression, decision tree and1333 Analysis period: 2015–2019 models, random forest is the method that1334 random forest.1335 Sample size: 506 yields the best results, as it has the lowest1336 error rate and performed well on both the1337 training and test data.1338 Its goal is to use machine learning models1339 for selecting hedonic variables and for1340 housing sale price models. Several models1341 Yoo et al. 2012 [35]1342 HPM, Linear Regression, Random were applied, and random forest resulted1343 Analysis period: 20001344 Forest and Cubist. in the best accuracy in terms of modeling,1345 Sample size: 44691346 with the potential to be useful for selecting1347 important variables for the hedonic price1348 equation.1349 A comparison of both models, including1350 the structural characteristics of the1351 property, envi-ronmental information, and1352 Ceh et al. 2018 [36]1353 neighborhood characteristics, was1354 Analysis period: 2008–2013 HPM and random forest.1355 conducted. The best predic-tions were1356 Sample size: 74071357 obtained with the random forest method,1358 which achieved a higher R2 and a1359 significantly lower error rate.1360 After using multiple machine learning1361 Tochaiwat and Pultawee,1362 techniques to analyze urban development1363 2024 [45]1364 Decision Tree, Random Forest and projects, it is found that combining models1365 Analysis period:1366 Gradient-Boosted Tree. yields better results providing more1367 2011/2014/2017/20211368 accurate and realistic statistics compared1369 Sample size: 591370 to the individual models on their own.1371 Urban Sci. 2025, 9, 32 17 of 201372137313741375 Table A1. Cont.13761377 Authors Compared Prediction Models Summary/Conclusions1378 It aims to analyze the impact of metro1379 stations and social capital on housing1380 prices. To accomplish this, it compares1381 eight machine learning methods to1382 Linear Regression, Decision Tree,1383 Paik et al. 2023 [53] provide more information for determining1384 Random Forest, LightGBM, Lasso1385 Analysis period: 2021–2022 housing prices. The LightGBM model has1386 Regression, Ridge Regression, Elastic1387 Sample size: 58,342 the smallest relative error between the1388 Net, and XGBoost.1389 actual and predicted values, and also1390 performed better in terms of absolute error1391 compared to the other models, making it1392 the most suitable model for this study.1393 To understand housing price fluctuations,1394 a model composed of two methodologies1395 is proposed as an experiment to test its1396 effectiveness. The result shows that1397 Rui and Liu, 2019 [54] Neural Network (Short-Term1398 combining the neural network with1399 Analysis period: 2010–2017 Memory), GA (Genetic Algorithm),1400 genetic algorithms successfully predicts1401 Sample size: 664 and Support Vector Regression.1402 housing prices with a better feature1403 selection process. However, a limitation is1404 that if the dataset is small, the1405 model weakens.1406 The housing price valuation study is1407 expanded by including regional clusters1408 such as proxim-ity to workplaces or1409 Kou et al. 2021 [55]1410 Neural Networks, XGBoost, and shopping centers. XGBoost presents the1411 Analysis period: 20181412 Support Vector Machine. highest R2 , and furthermore, when1413 Sample size: 158,8881414 economic clusters are added, they make1415 the fit more accurate than all the1416 traditional features.1417 It aims to analyze the amenities of1418 neighborhoods and condominium prices1419 in the city. Both the popularity and1420 Taecharungroj, 2021 [56]1421 availability of services drive1422 Analysis period: 2021 Random forest and XGBoost.1423 condominium prices in non-linear ways.1424 Sample size: 152,5121425 The XGBoost model shows a higher level1426 of fit (R2 ), making it perform better than1427 random forest.1428 Four machine learning models are1429 developed to predict housing prices.1430 Additionally, a hybrid model is created for1431 the same purpose. Neural networks1432 Chou et al. 2022 [27] Artificial Neural Networks, Linear1433 achieved the best perfor-mance in R2 ,1434 Analysis period: 2013–2017 Regression, Regression Tree, Support1435 RMSE, MAE, and MAPE. However, the1436 Sample size: 209,402 Vector Machine, and Hybrid Model.1437 hybrid model, by combining three models,1438 demonstrates greater precision than each1439 individual model, as it leverages the1440 advantages of each.1441 Urban Sci. 2025, 9, 32 18 of 201442144314441445 Table A1. Cont.14461447 Authors Compared Prediction Models Summary/Conclusions1448 It aims to investigate the determinants1449 considered in the models when valuing1450 housing prices through the application of1451 Iban, 2022 [57] four machine learning methods. In this1452 Random forest, XGBoost, LightGBM1453 Analysis period: 2021 study, the Gradient Boosting model1454 and Gradient Boosting.1455 Sample size: 1002 achieved the highest R2 score. However,1456 the XGBoost model presented the lowest1457 MAPE and RMSE values, indicating1458 better performance.1459 The goal is to obtain the implicit prices of1460 housing characteristics by comparing the1461 two models being analyzed. Neural1462 Núñez-Tabales et al. 2012 [30]1463 HPM and Artificial networks demonstrate greater predictive1464 Analysis period: 20061465 Neural Networks. power and more satisfactory results due to1466 Sample size: 28881467 a higher degree of fit (R2 ) and lower1468 RMSE, MAE, and residual standard1469 deviation rates.1470 By using HPM and several machine1471 learning models, the aim is to determine1472 K-Nearest Neighbors, Decision Tree,1473 which of them provides the most1474 Rico-Juan and Taltavull, Random Forest, AdaBoost, XGBoost,1475 significant results in predicting housing1476 2021 [12] CatBoost, Artificial Neural Networks1477 prices. After analyzing all the models, the1478 Analysis period: 2004–2012 (Multilayer), Linear Regression,1479 random forest model proves to be the1480 Sample size: 392,412 Ridge Regression, Lasso Regression,1481 most suitable for the task, showing a1482 and HPM.1483 higher degree of fit and the lowest1484 error rates.1485 Source: own elaboration.148614871488References14891. Mark, J.; Kim, K. Segmentation of the housing market and its determinants: Seoul and its neighbouring new towns in Korea.1490 Aust. Geogr. 2007, 36, 221–232.14912. Wang, G. The effect of environment on housing prices: Evidence from the Google Street View. J. Forecast. 2023, 42, 288–311.1492 [CrossRef]14933. Bartholomew, K.; Edwing, R. Hedonic price effects of pedestrian- and transit-oriented development. J. Plan. Lit. 2011, 26, 18–34.1494 [CrossRef]14954. Howard, G.; Liebersohn, J. Regional divergence and house prices. Rev. Econ. Dyn. 2023, 49, 312–350. [CrossRef]14965. Jayantha, W.M.; Oladinrin, O.T. Bibliometric analysis of hedonic price model using CiteSpace. Int. J. Hous. Mark. Anal. 2019, 13,1497 357–371. [CrossRef]14986. Xiao, Y. Urban Morphology and Housing Market; Springer: Berlin/Heidelberg, Germany, 2017.14997. Lundberg, S.M.; Lee, S.-I. A unified approach to interpreting model predictions. In Proceedings of the 31st International1500 Conference on Neural Information Processing, Long Beach, CA, USA, 4–9 December 2017; pp. 4766–4777.15018. Fan, C.; Cui, Z.; Zhong, X. House prices prediction with machine learning algorithms. In Proceedings of the ICMLC’18: 2018 10th1502 International Conference on Machine Learning and Computing, Macau, China, 26–28 February 2018; pp. 6–10.15039. Rouhiaine, L. Inteligencia Artificial: 101 Cosas Que Debes Saber Hoy Sobre Nuestro Future; Editorial Planeta S.A.: Barcelona,1504 Spain, 2018.150510. Barzegar, R.; Adamowski, J.; Moghaddam, A.A. Application of wavelet-artificial intelligence hybrid models for water quality1506 prediction: A case study in Aji-Chai River, Iran. Stoch. Environ. Res. Risk Assess. 2016, 30, 1797–1819. [CrossRef]150711. Glaeser, E.L.; Kominers, S.D.; Luca, M.; Naik, N. Big data and big cities: The promises and limitations of improved measures of1508 urban life. Econ. Inq. 2016, 56, 114–137. [CrossRef]150912. Rico-Juan, J.R.; de la Paz, P.T. Machine learning with explainability or spatial hedonics tools? An analysis of the asking prices in1510 the housing market in Alicante, Spain. Expert Syst. Appl. 2021, 171, 114590. [CrossRef]151113. Goodman, A.C. Andrew Court and the invention of hedonic price analysis. J. Urban Econ. 1998, 44, 291–298. [CrossRef]1512 Urban Sci. 2025, 9, 32 19 of 20151315141515151614. Rosen, S. Hedonic prices and implicit markets: Product differentiation in pure competition. J. Political Econ. 1974, 82, 34–55.1517 [CrossRef]151815. Diewert, W.E.; de Haan, J.; Hendricks, R. Hedonic regressions and the decomposition of a house price index into land and1519 structure components. Econom. Rev. 2011, 34, 106–126. [CrossRef]152016. Hwang, M.; Quigley, J. Economic fundamentals in local housing markets: Evidence from U.S. metropolitan regions. J. Reg. Sci.1521 2006, 46, 425–453. [CrossRef]152217. Mendoza, N.J. La Red de Hopfield. Rev. Inf. Tecnol. Y Soc. 2009, 35.152318. Shinde, P.P.; Shah, S. A review of machine learning and deep learning applications. In Proceedings of the International Fourth1524 Conference on Computing Communication Control and Automation (ICCUBEA), Pune, India, 16–18 August 2018; pp. 1–6.152519. Cortes, C.; Vapnik, V. Support-vector networks. Mach. Learn. 1995, 20, 273–297. [CrossRef]152620. Angra, S.; Ahuja, S. Machine learning and its application: A review. In Proceedings of the International Conference on Big Data1527 Analytics and Computational Intelligence (ICBDAC), Chirala, India, 23–25 March 2017; pp. 57–60.152821. Song, Y.; Ma, X. Exploration of intelligent housing price forecasting based on the anchoring effect. Neural Comput. Appl. 2024, 36,1529 2201–2214. [CrossRef]153022. Arias-Odón, F. Investigación documental, investigación bibliométrica y revisiones sistemáticas. Rev. Electrónica De Humanidades1531 Educ. Y Comun. Soc. 2023, 31, 9–28.153223. Manterola, C.; Astudillo, P.; Arias, E.; Claros, N.; Mincir, G. Revisiones sistemáticas de la literatura. Qué se debe saber acerca de1533 ellas. Cirugía Española 2013, 91, 149–155. [CrossRef]153424. Page, M.J.; McKenzie, J.E.; Bossuyt, P.M.; Boutron, I.; Hoffmann, T.C.; Mulrow, C.D.; Shamseer, L.; Tetzlaff, J.M.; Akl, E.A.;1535 Brennan, S.E.; et al. Declaración PRISMA 2020: Una guía actualizada para la publicación de revisiones sistemáticas. Rev. Española1536 De Cardiol. 2021, 74, 790–799. [CrossRef]153725. Beltrán, Ó.A. Revisiones sistemáticas de la literatura. Rev. Colomb. De Gastroenterol. 2005, 20, 60–69.153826. Sarkis-Onofre, R.; Catalá-López, F.; Aromataris, E.; Lockwood, C. How to properly use the PRISMA statement. Syst. Rev. 2021,1539 10, 117. [CrossRef]154027. Chou, J.S.; Fleshman, D.B.; Truong, D.N. Comparison of machine learning models to provide preliminary forecasts of real estate1541 prices. J. Hous. Built Environ. 2022, 37, 2079–2114. [CrossRef]154228. Rigatti, S.J. Random forest. J. Insur. Med. 2017, 47, 31–39. [CrossRef] [PubMed]154329. Shmueli, G. To explain or to predict? Stat. Sci. 2010, 25, 289–310. [CrossRef]154430. Tabales, J.M.; y Ocerin, J.M.; Villamandos, N.C.; Carmona, F.J. Obtención de precios implícitos para atributos determinantes en la1545 valoración de una vivienda. Rev. Int. De Adm. Y Finanz. 2012, 5, 41–54.154631. Chen, M.; Liu, Y.; Arribas-Bel, D.; Singleton, A. Assessing the value of user-generated images of urban surroundings for house1547 price estimation. Landsc. Urban Plan. 2022, 226, 104486. [CrossRef]154832. Hong, J.; Choi, H.; Kim, W.S. A house price valuation based on the random forest approach: The mass appraisal of residential1549 property in South Korea. Int. J. Strateg. Prop. Manag. 2020, 24, 1–13. [CrossRef]155033. Chen, J.-H.; Ong, C.F.; Zheng, L.; Hsu, S.-C. Forecasting spatial dynamics of the housing market using Support Vector Machine.1551 Int. J. Strateg. Prop. Manag. 2017, 21, 273–283. [CrossRef]155234. Selim, H. Determinants of house prices in Turkey: Hedonic regression versus artificial neural network. Expert Syst. Appl. 2009, 36,1553 2843–2852. [CrossRef]155435. Yoo, S.; Im, J.; Wagner, J.E. Variable selection for hedonic model using machine learning approaches: A case study in Onondaga1555 County, NY. Landsc. Urban Plan. 2012, 107, 293–306. [CrossRef]155636. Čeh, M.; Kilibarda, M.; Lisec, A.; Bajat, B. Estimating the performance of random forest versus multiple regression for predicting1557 prices of the apartments. Int. J. Geo-Inf. 2018, 7, 168. [CrossRef]155837. Schläpfer, F.; Waltert, F.; Segura, L.; Kienast, F. Valuation of landscape amenities: A hedonic pricing analysis of housing rents in1559 urban, suburban, and periurban Switzerland. Landsc. Urban Plan. 2015, 141, 24–40. [CrossRef]156038. Wang, D.; Li, V.J. Mass appraisal models of real estate in the 21st century: A systematic literature review. Sustainability 2019,1561 11, 7006. [CrossRef]156239. Ligus, M.; Peternek, P. Impacts of Urban Environmental Attributes on Residential Housing Price in Warsaw (Poland): Spatial Hedonic1563 Analysis of City Districts; Springer: Berlin/Heidelberg, Germany, 2017; pp. 155–164.156440. Crespo, R.; Grêt-Regamey, A. Local hedonic house-price modelling for urban planners: Advantages of using local regression1565 techniques. Environ. Plan. B Plan. Des. 2013, 40, 664–682. [CrossRef]156641. Afonso, B.; Melo, L.; Oliveira, W.; Sousa, S.; Berton, L. Housing prices prediction with a deep learning and random forest1567 ensemble. In Proceedings of the Anais do Encontro Nacional de Inteligência Artificial e Computacional (ENIAC), Salvador, Brasil,1568 15–18 October 2019; pp. 389–400.156942. Hoxha, V. Comparative analysis of machine learning models in predicting housing prices: A case study of Prishtina’s real estate1570 market. Int. J. Hous. Mark. Anal. 2024. [CrossRef]1571 Urban Sci. 2025, 9, 32 20 of 20157215731574157543. Ho, W.K.O.; Tang, B.-S.; Wong, S.W. Predicting property prices with machine learning algorithms. J. Prop. Res. 2020, 38, 48–70.1576 [CrossRef]157744. Begum, A.; Kheya, N.J.; Rahman, M.Z. Housing price prediction with machine learning. Int. J. Innov. Technol. Explor. Eng. 2022,1578 11, 42–46. [CrossRef]157945. Tochaiwat, K.; Pultawee, P. House type specification for housing development project using machine learning techniques: A1580 study from Bangkok Metropolitan Region, Thailand. J. Environ. Plan. 2024, 23, 403. [CrossRef]158146. Adetunji, A.B.; Akande, O.N.; Ajala, F.A.; Oyewo, O.; Akande, Y.F.; Oluwadara, G. House price prediction using random forest1582 machine learning technique. Procedia Comput. Sci. 2022, 199, 806–813. [CrossRef]158347. Soltani, A.; Heydari, M.; Aghaei, F.; Pettit, C.J. Housing price prediction incorporating spatio-temporal dependency into machine1584 learning algorithms. Cities 2022, 131, 103941. [CrossRef]158548. Chiarazzo, V.; Caggiani, L.; Marinelli, M.; Ottomanelli, M. A neural network-based model for real estate price estimation1586 considering environmental quality of property location. Transp. Res. Procedia 2014, 3, 810–817. [CrossRef]158749. Azadeh, A.; Sheikhalishahi, M.; Boostani, A. A flexible neuro-fuzzy approach for improvement of seasonal housing price1588 estimation in uncertain and non-linear environments. South Afr. J. Econ. 2014, 82, 567–582. [CrossRef]158950. Chaturvedi, V.; de Vries, W.T. Machine learning algorithms for urban land use planning: A review. Urban Sci. 2021, 5, 68.1590 [CrossRef]159151. Chen, Y.; Jiao, J.; Farahi, A. Disparities in affecting factors of housing price: A machine learning approach to the effects of housing1592 status, public transit and density factors on single-family housing price. Cities 2023, 140, 104432. [CrossRef]159352. Lahmiri, S.; Bekiros, S.; Avdoulas, C. A comparative assessment of machine learning methods for predicting housing prices using1594 Bayesian optimization. Decis. Anal. J. 2023, 6, 100166. [CrossRef]159553. Paik, J.; Baek, S.-J.; Kim, J.-W.; Ko, K. Influence of social overhead capital facilities on housing prices using machine learning.1596 Appl. Sci. 2023, 13, 10732. [CrossRef]159754. Rui, L.; Liu, L. Predicting housing price in China based on long short-term memory incorporating modified genetic algorithm.1598 Soft Comput. 2019, 23, 11829–11838.159955. Kou, J.; Du, J.; Fu, X.; Zhang, G.Z.; Wang, H.; Zhang, Y. The Effect of Regional Economic Clusters on Housing Price; Lecture Notes in1600 Artificial Intelligence; Springer: Berlin/Heidelberg, Germany, 2021.160156. Taecharungroj, V. Google Maps amenities and condominium prices: Investigating the effects and relationships using machine1602 learning. Habitat Int. 2021, 118, 102409. [CrossRef]160357. Iban, M.C. An explainable model for the mass appraisal of residences: The application of tree-based machine learning algorithms1604 and interpretation of value determinants. Habitat Int. 2022, 128, 102409. [CrossRef]160516061607Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual1608author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to1609people or property resulting from any ideas, methods, instructions or products referred to in the content.1610