Predicting property prices with machine learning algorithms

TLDR

The study applies support vector machine, random forest, and gradient boosting machine algorithms to predict property prices. These models were trained and evaluated on a dataset of roughly 40,000 Hong Kong housing transactions spanning more than 18 years, and their predictive performances were compared. Random forest and gradient boosting achieved superior predictive accuracy over SVM, as shown by lower MSE, RMSE, and MAPE, though SVM still provides reasonably accurate predictions within tight time constraints, indicating machine learning is a promising alternative for property valuation.

Abstract

This study uses three machine learning algorithms including, support vector machine (SVM), random forest (RF) and gradient boosting machine (GBM) in the appraisal of property prices. It applies these methods to examine a data sample of about 40,000 housing transactions in a period of over 18 years in Hong Kong, and then compares the results of these algorithms. In terms of predictive power, RF and GBM have achieved better performance when compared to SVM. The three performance metrics including mean squared error (MSE), root mean squared error (RMSE) and mean absolute percentage error (MAPE) associated with these two algorithms also unambiguously outperform those of SVM. However, our study has found that SVM is still a useful algorithm in data fitting because it can produce reasonably accurate predictions within a tight time constraint. Our conclusion is that machine learning offers a promising, alternative technique in property valuation and appraisal research especially in relation to property price prediction.