Türkiye’deki Konut Fiyatlarının Tahmininde Hedonik Regresyon Yöntemi ile Yapay Sinir Ağlarının Karşılaştırılması

Abstract

Owner-occupied housing is both a place to live and also the most important asset in many households’ portfolio. Accurately predicting of house prices is therefore of great interest to the general public. This paper aims to compare the housing price prediction accuracies of Hedonic Model (HM) and Artificial Neural Networks (ANNs). In order to achieve this aim, two techniques’ prediction results were compared by using four performance criteria: RMSE, MAE, MAD, and Theil’s U statistic. This study uses the HM and ANNs to empirically determine the house prices in Turkey. HM is the standard technique for modeling the behavior of house prices over the past three decades and is based on micro economic theory. The non-linear relationship between house price and its determinants can be modeled by an ANN, so it is employed in this paper as an alternative method. Empirical results revealed that ANNs performed better than HM in house price predictions, indicating that ANNs could be useful for prediction of house prices. More clearly, the performance criteria from the ANNs are smaller than those from the HM by roughly 60-90%. For instance, the ANN model has about 77 percent lower RMSE, 91 percent lower MAE, 64 percent lower MAD, and 77 percent lower Theil’s U statistic than those of the HM.