House Prices Prediction Using Statistics with Machine Learning

Abstract

After the housing crisis in 2009 that affected the global economy and the bubble that burst, researchers began to focus on how to estimate house prices. In the United States, for instance, they adopted the hedonic price index (HPI) method in estimating house prices. After Ames house pricing dataset was released, which contains houses data from 2006 to 2010, and detailed features that help in studying the estimation of house prices. In this paper, we suggest that House prices are determined by many features such as area, utilities, house style, location, age, grade living area, number of bedrooms, garage, and so on. Statistical methods were applied with two models which are multiple and stepwise linear regression, also, two machine learning algorithms which are LASSO and XGBoost regression. The accuracy of prediction was evaluated by the root mean square error (RMSE). XGBoost with 25 features, 0.973 R2, and 0.027 RMSE is the Best model. LASSO has helped in feature selection for XGBoost.