Machine Learning Models for Real Estate Appraisal Constructed Using Spline Trend Functions

Abstract

The paper presents methods of modeling the real estate market using trend functions reflecting changes in real estate prices over time. Real estate transaction prices that are used to create data-driven valuation models must be updated in line with the trend of their change. The primary purpose of the first part of the study was to examine the extent to which splines are suitable for the trend function compared to polynomials of the degree from 1 to 6. In turn, the second part was to compare the performance of prediction models built on the basis of updated data with various trend functions: splines and polynomials. The experiments were conducted using real data on purchase and sale transactions of residential premises concluded in one of the Polish cities. Four machine learning algorithms implemented in the Python environment were used to generate property valuation models. Statistical analysis of the results was carried out using non-parametric Friedman and Wilcoxon tests. The study showed the usefulness of applying splines to model trend functions.