Modelling the Unit Property Prices in Türkiye Employing Deep Learning Methods

Abstract

The real estate market in Türkiye has witnessed significant growth and complexity in recent years necessitating advanced modelling techniques for accurate property price predictions. This paper presents a comprehensive study on the application of deep learning networks to model unit property prices across Türkiye, with a specific emphasis on its three major cities: Istanbul, Ankara, and Izmir. The unit property prices in the overall country and its three major cities are taken from the official resources and then the square meter unit property prices in USD are obtained. Then, a deep learning network is developed in Python programming language which accepts the lagged values of the unit property prices as input values. Then, the deep learning network is trained using the datasets separately for Türkiye and the three major cities. The loss curves of the deep learning network show a rapid convergence indicating the suitability of the developed deep learning network for the unit property price modelling. The actual and the modelled data are plotted indicating the accuracy of the developed model. Finally, the performance indicators of the developed deep learning model namely the coefficient of determination, mean absolute error, root mean square error and the mean absolute percentage error are computed verifying the high accuracy with R2 values greater than 0.95 for all four modelling cases. The alternative utilization opportunities of the proposed deep learning model are also discussed in the conclusions section.