Comparing modelling performance and evaluating differences of feature importance on defined geographical appraisal zones for mass real estate appraisal

ABSTRACT

The features influencing real estate value in different residential areas and cities are important for spatial economic analysis besides high appraisal accuracy. In this study, a methodology was developed for computer-assisted mass real estate appraisal with a case study implemented through the use of big geographical datasets including 121 features and around 200,000 samples of real estate in Istanbul and Kocaeli (Turkey). Prediction models using the random forest technique were developed for five appraisal zones determined with spatially constrained multivariate clustering. With machine learning and mass appraisal metrics, modelling performance improves in appraisal zones with a lower standard deviation expressing real estate value in neighbourhoods. Since importance levels and ranks of features vary in zones, the mass appraisal should be done with a sufficient number of features.

KEYWORDS:

• computer-assisted mass appraisal
• geographical appraisal zone
• machine learning
• random forest
• spatially constrained multivariate clustering analysis

JEL:

• C8
• R14
• R21

ACKNOWLEDGEMENTS

We thank anonymous reviewers and the editor, for their supportive positive comments and suggestions that significantly improved the final version of this paper.

DISCLOSURE STATEMENT

No potential conflict of interest was reported by the authors.