Real estate valuation based on big data

Abstract

The paper considers the application of the web scrapping and machine learning algorithms for the assessment of the real estate price on the secondary housing market in Moscow. For this, we collect and process the data from the CIAN website and the data from “Reforma GKH”. To evaluate real estate objects, we consider such machine learning algorithms as Elastic Net, Random Forest and Gradient Boosting. We also apply Shapley vector-based approach to interpret the results of the black-box algorithms. The results suggest that the use of black-box algorithms in assessing the price of apartments on the Moscow secondary housing market allows to obtain more accurate price estimates both for different price segments and for the sample as a whole. At the same time, Gradient Boosting has demonstrated the best accuracy among other algorithms. Interpretation based on the Shapley vector shows that the total area, year of construction, ceiling height, renovation, as well as monolithic construction technology had a positive effect on the price. The price is negatively affected by the number of floors in the house, the possibility of mortgage and lack of repairs. Developed methodology can be applied in real estate insurance, mortgage, determination of cadastral value of real estate and others.