Real estate valuation using machine learning techniques

Abstract

In this work, the methods of artificial intelligence are analyzed in order to perform a more accurate value prediction of the apartments sold in Vilnius city and district. The work uses publicly available information about apartments put for sale on Aruodas.lt, which is collected in an automated way. The information that is collected consists of a text – description of the apartment for sale, photos – photos placed in the ad, and general information provided in the ad – price, location, apartment size, various features of the apartment and more. A constant problem in real estate valuation is the overvaluation of low-value objects and/or underestimation of high-value objects. When dealing with regression problems, we often have data on most average objects, but never enough of cheap and luxurious ones. For this reason, estimating the value of most properties is easier than cheap or expensive. However, as the methods of artificial intelligence evolve and information becomes more and more available, our ability to better value this type of housing increases. It is hoped that the information contained in the photos and text will allow for a better forecast of the value of housing by better valuing both cheap and expensive apartments. In the first part of the work, a review of the literature is performed where other authors have examined the possibilities of using artificial intelligence to predict the value of housing. The second part describes the research methods that will be applied in the work and presents the information gathering strategy. In the third part, a study is conducted, during which feature engineering is performed first, followed by model training and optimization. Finally, the results of the best model with 13.74 MAPE and 33,307 RMSE are presented along with the conclusions of the work.