UNDERSTANDING HDB RESALE PRICES IN SINGAPORE: A MACHINE LEARNING & ECONOMETRICS APPROACH.

Abstract

This study evaluates the efficacy of machine learning techniques using Housing Development Board (HDB) public housing resale prices in Singapore. More specifically, in the areas of prediction, variable selection, and hedonic price function estimation. Based on the evaluation done in this study, machine learning methods provide improvements over the ordinary least squares and a means to distinguish factors that could influence public housing prices. Nonlinear algorithms such as boosting, neural networks, hybrids, and ensembles were among the methods with the best prediction performance. Moreover, post-LASSO with covariates expanded via feature engineering could identify several significant interactions in the hedonic price function and had a better model fit than the hedonic model estimated by the ordinary least squares.