Artificial Neural Networks vs Spatial Regression Approach in Property Valuation

Abstract

The purpose of this paper is to compare two approaches applied in property
valuation: artificial neural networks and spatial regression. Despite the fact
that artificial neural networks are often the first choice for modeling in the
big data era, spatial econometrics methods offer incorporation of information
on dependences between multiple objects in the studied space. Although this
dependency structure can be incorporated into artificial neural network via
feature engineering, this study is focused on abilities of reproducing it with
machine learning method from crude coordinate data. The research is based on
the database of 18,166 property sale transactions in Warsaw, Poland. According
to this study, such volume of data does not allow artificial neural networks
to compete in reflecting spatial dependence structure with spatial regression
models.