Making Housing Predictions Using ML Without Live Market Data

Abstract

This paper explores the importance of machine learning and the benefits machine learning can bring to the housing industry. The housing industry makes up an average of more than 15% of the US’s yearly GDP and plays a significant role in the US economy. Additionally, the housing industry, specifically house prices, determines the environment each person grows up in by filtering out those who are unable to afford the costs of living in any given area. Having accurate house price information on hand can save many hours spent looking for affordable housing. After a brief introduction of Zillow, a house listing website that also contains a feature that provides highly accurate predictions of its users’ house prices, the central question, “How high a degree of accuracy can house prices be predicted without information regarding the state of the current economy or house market?” is explored. While Zillow uses data such as market trends and past selling prices, only information regarding the physical traits of the house, such as its geographical location and its room count, are considered within this paper (Note that the scaling factor for many variables is messed up, meaning that despite numbers that don’t intuitively seem correct, the relationship between each pair of data points remains unaffected). After visually identifying correlations between house value and the other variables with simple scatter plots, the relationships between different variables were further explored with different regression models, and a neural network was made based off of the previous exploration.