Analysis of Housing Market to Predict Home Price

Predicting the house market and home prices is still a challenging task. The price of a house depends on several factors and it is important to find the key factors that affect the price of a house in a particular location. In this study, we conducted an experiment to observe how certain variables affect housing prices. Principal component analysis (PCA) and a correlation heatmap were developed to document which variables had the highest correlation with price. At first, these variables were isolated for further development. Using the scikit-learn and XGBoost libraries for Python several models were developed for predicting housing prices within the area of Kings County, California, United States. From these models it was determined that using the full range of variables, rather than the most highly correlated variables, yielded the most accurate results. Out of the three regression models developed, the XGBoost model yielded the most accurate predictions compared to linear regression and random forest.

I. Introduction

Data mining and machine learning are crucial when it comes to identifying useful knowledge and patterns within large sets of data. One of the largest data sets that we do not normally think about but hear so much about is the housing market. One big thing to note is the past history of the housing market where during the mid-2000s the market began to fall and only began to recover after 2012. The volatility of the market has proved to be much more than just economic changes involving aspects such as mortgage rates. This seems to be one of the factors that is most talked about, especially within news outlets and other sources of information. However, going beyond the surface with a fresh set of eyes is necessary in order to find a modern solution to a very modern problem.