Machine learning as a tool for improved housing price prediction : the applicability of machine learning in housing price prediction and the economic implications of improvement to prediction accuracy

Abstract

This thesis investigates whether non-linear machine learning algorithms can produce more accurate predictions of Norwegian housing prices compared to linear regression models. We find that the non-linear XGBoost algorithm increases out-of-sample prediction accuracy by 8.5% in terms of Root Mean Squared Error compared to the linear model used by Statistics Norway. Using additional property-specific and macroeconomic variables such as coordinates, common debt, story, inflation rate and interest rate, we find that a non-linear Stacked Regression model improves out-of-sample prediction accuracy by 39.52% in terms of Root Mean Squared Error compared to a linear model.