Machine learning in house price prediction
No Thumbnail Available
URL
Journal Title
Journal ISSN
Volume Title
School of Business |
Master's thesis
Authors
Date
2022
Department
Major/Subject
Mcode
Degree programme
Information and Service Management (ISM)
Language
en
Pages
78
Series
Abstract
During the last decades, housing prices have been a frequent topic in economic discussions. Due to the interconnectedness and the magnitude of the housing markets of individual national economies, shocks in the housing markets have had a significant impact in economies around the world, both in economic booms as well as in times of recession. As everyone is affected by the price shocks of the housing markets, precise pricing of individual houses and apartments plays an important role in the efficiency of the housing markets. As the theory of efficient markets suggests, all participants should have all available information regarding the prices. In order to make the Finnish housing markets more efficient, an accurate machine learning based pricing model could be of help. This thesis will dive deep into the pricing of houses and, more precisely, into the utilisation of different machine learning methods for house price prediction. While the previous research has covered different variations of classical regression models in price predictions and tried to uncover the underlying factors impacting the prices, machine learning models have not yet been exhaustively researched in Finland. Three of the most popular machine learning methods, according to earlier studies, were compared in this study with the best model being chosen according to RMSE, MAE and r-squared evaluation metrics. From the three most popular methods, Random Forests model performed the best in out-of-sample predictions. The results support earlier studies and real-life use cases by confirming that the utilisation of machine learning methods in house price predictions is feasible. The model is able to predict the prices of apartments in the Helsinki Metropolitan area with an average error of approximately €404 in price per square metre which corresponds to an average percentual error of 7.8%. The most important factors influencing the prices according to this study were multiple spatial variables, the building year, the size of the apartment and elevator. These results are also in line with earlier research. As the constructed model is robust and accurate enough, it could be used as a pricing tool for any participant in the housing markets. This paper fills research gaps in the field of house price prediction in Finland and the newly constructed model could help with the problem of information asymmetry in the housing markets.Description
Thesis advisor
Liao, ZhiqiangKeywords
housing market, house prices, predictive analytics, machine learning, random forests