Linear regression

Random forest

Engineering

Statistics

Cartography

Data mining

Civil engineering

Computer science

Real estate

Business

Cadastre

Mathematics

Geography

Residential area

Data science

Artificial intelligence

Finance

Categorical variable

Geospatial analysis

Machine learning

Land Use Policy

Mass appraisal is a complex matter because it depends on several categorical and continuously changing or constant parameters. In addition, development of new assessment approaches for mass appraisal of real estate properties in highly complex urban environments is desirable. The advancements in geospatial technologies and machine learning algorithms open up new horizons. For this reason, the purpose of the present study is to compare one conventional stepwise linear multiple regression (MRA) and one more automated machine learning approach, random forest (RF), for mass appraisal in an urban residential area where commercial properties are also available. A part of Mamak District, Ankara, Turkey is selected as the study area since the property values are diverse and representative. Additionally, the district has a complex and developing urban structure. The data employed in the study were compiled under a cadastral modernization project of General Directorate of the Land Registry and Cadastre of Turkey (GDLRC) and were based on the reports of licensed experts (∼50 %), court reports (∼20 %), field surveys, or a combined analysis of all. Consequently, the data used in the study has a high level of confidence. The initial set of parameters used in both methods reflect the most frequently observed characteristics of the real estate properties in the study area that are also effective on the values. The stepwise MRA required manual adjustments of the final parameter set by the expert, whereas RF eliminated unusable parameters fully automatically. The method performance was assessed by using a subset of the training data as a random test. According to the accuracy assessment results, the RF (Adjusted R² 0.734; the total variance explained from the model) slightly outperforms the MRA (Adjusted R² 0.696) where the optimal parameters were set by the human expert. Finally, the results exhibited are promising for quick assessment of mass appraisal and a comprehensive discussion is presented in the study.

A mass appraisal assessment study using machine learning based on multiple regression and random forest