APPLICATION OF MACHINE LEARNING MODELS FOR PREDICTING PRICES ON THE REAL ESTATE MARKET

Abstract

The work shows the results of using machine learning to forecast price changes in the real estate market. Economic models and factors affecting price formation are analyzed. This data was used as a basis for building a machine learning model. Special attention is focused on the selection of basic data for building such a model. Various types of regression models, which can be implemented in appropriate software environments, have been analyzed. As a result, a statistical model was created for predicting housing prices using linear regression. It is used to determine trends in price changes on the real estate market in the medium and long term. All libraries used in the development of statistical models are examined in detail, their advantages and disadvantages are analyzed. The following steps of model creation were considered and implemented: import of libraries and modules, developed data reading from the dataset, data analysis, cleaning and average statistical evaluation. As a result, linear regression was used to analyze an array of data obtained from an open resource – the real estate sales and rental site Zillow. The real estate objects described on this resource relate to the city of Seattle. Clustering was used according to the main parameters of real estate objects, in particular, area, location, age. In order to evaluate the ratio of various characteristics of the real estate object, 3D modeling was carried out using the Axes3D designer. Based on the analysis, data visualization was performed using various libraries. The analyzed data is displayed on a map using Folium. Data clustering and testing on real data were implemented, which showed quite good price forecast results. The obtained modeling result was checked on real estate objects and it was established that the accuracy of the model is 76%. Such a high result indicates the correctness of its construction and the rationality of using software solutions for its implementation. In the future, it can be used to analyze similar data sets in this field.