A new submarket approach using distances to transit lines for the prediction of real estate prices

Abstract

Prediction of a price of a real estate has been one of the trend topics in recent years. There are many studies conducted on both prediction of a value of a real estate or the a ecting parameters. In this study, the prediction of a real estate price using submarket near transit lines is studied on two neighbor counties in Istanbul, Beylikduzu and Esenyurt. So, the data should be analyzed into parts to investigate altered dynamics of different districts, which is also called submartket analysis. After the whole data of real estates in these counties are collected and analyzed, the data are divided into three parts (Esenyurt, Beylikduzu and transition zone) and analyzed in order to investigate altered dynamics of different districts. A total of 3487 real estate data with one dependent variable and 13 independent variables collected from aforementioned districts are analyzed with two machine learning (Multiple Linear Regression (MLR) and Spatial Auto Regression (SAR) and one deep learning tool (MultiLayer Perceptron (MLP)). According to the results of the both whole data and submarket analysis, Spatial Auto Regressive model is superior to the others in a metric of R-squared. Moreover, with the submarket analysis, prediction power of all the algorithms (MLR, SAR, and MLP) are signi cantly increased. Significant independent variables of each model differ from each other so that it can be concluded that submarket analysis in real estate prediction is improving the prediction models and showing different dynamics of each specific district of a county.