Machine Learning-Based Feature Mapping for Enhanced Understanding of the Housing Market

TY - GEN

T1 - Machine Learning-Based Feature Mapping for Enhanced Understanding of the Housing Market

AU - Lystbæk, Michael Sahl

AU - Srirajan, Tharsika Pakeerathan

N1 - Publisher Copyright: © The Author(s), under exclusive license to Springer Nature Switzerland AG 2024.

PY - 2024

Y1 - 2024

N2 - The housing market is impacted by a variety of parameters which gives a complexity that is difficult to analyze with traditional statistical approaches due to the large number of interdependent variables that the market data provides. In this study, ML techniques are utilized to provide a deeper understanding of the Danish housing market based on a dataset of sales cases provided by a leading Danish real estate agency. We propose an extreme gradient boosting model for sales price regression, and we propose using feature importance techniques to provide insight into important parameters in the national housing market. The regression model trained for sales price with grid search cross-validation for parameter optimization achieves an R2 accuracy of 0.84, an MAE of DKK 433,824, and an RMSE of DKK 675,817. Permutation-based feature importance defines the most impactful parameters for the sales price regression where the four features with the highest impacts are: 1. GisX (West/East location), 2. GisY (North/South location), 3. Building area, 4. Construction year. The results for geographical distribution regarding price, building area, and plot area are illustrated with 2D partial dependence plots of geographical distributions to enhance the understanding of market trends.

AB - The housing market is impacted by a variety of parameters which gives a complexity that is difficult to analyze with traditional statistical approaches due to the large number of interdependent variables that the market data provides. In this study, ML techniques are utilized to provide a deeper understanding of the Danish housing market based on a dataset of sales cases provided by a leading Danish real estate agency. We propose an extreme gradient boosting model for sales price regression, and we propose using feature importance techniques to provide insight into important parameters in the national housing market. The regression model trained for sales price with grid search cross-validation for parameter optimization achieves an R2 accuracy of 0.84, an MAE of DKK 433,824, and an RMSE of DKK 675,817. Permutation-based feature importance defines the most impactful parameters for the sales price regression where the four features with the highest impacts are: 1. GisX (West/East location), 2. GisY (North/South location), 3. Building area, 4. Construction year. The results for geographical distribution regarding price, building area, and plot area are illustrated with 2D partial dependence plots of geographical distributions to enhance the understanding of market trends.

KW - Feature Importance

KW - Gradient Boosting Regression

KW - Housing Market Pricing

KW - Machine Learning

KW - Partial Dependence

UR - http://www.scopus.com/inward/record.url?scp=85198979257&partnerID=8YFLogxK

U2 - 10.1007/978-3-031-62495-7_40

DO - 10.1007/978-3-031-62495-7_40

M3 - Article in proceedings

AN - SCOPUS:85198979257

SN - 978-3-031-62494-0

T3 - Communications in Computer and Information Science

SP - 530

EP - 543

BT - Engineering Applications of Neural Networks

A2 - Iliadis, Lazaros

A2 - Maglogiannis, Ilias

A2 - Papaleonidas, Antonios

A2 - Pimenidis, Elias

A2 - Jayne, Chrisina

PB - Springer Science and Business Media Deutschland GmbH

T2 - 25th International Conference on Engineering Applications of Neural Networks, EANN 2024

Y2 - 27 June 2024 through 30 June 2024

ER -