Decision Tree and Boosting Techniques in Artificial Intelligence Based Automated Valuation Models (AI-AVM)

Abstract

This paper develops an artificial intelligence-based automated valuation model (AI-AVM) using the decision tree and the boosting techniques to predict residential property prices in Singapore. We use more than 300,000 property transaction data from Singapore’s private residential property market for the period from 1995 to 2017 for the training of the AI-AVM models. The two tree-based AI-AVM models show superior performance over the traditional multiple regression analysis (MRA) model when predicting the property prices. We also extend the application of the AI-AVM to more homogenous public housing prices in Singapore, and the predictive performance remains robust. The boosting AI-AVM models that allow for inter-dependence structure in the decision trees is the best model that explains more than 88% of the variance in both private and public housing prices and keep the prediction errors to less than 6% for HDB and 9% for the private market. When subject the AI-AVM to the out-of-sample forecasting using the 2017-2019 testing property sale samples, the prediction errors remain within a narrow range of between 5% and 9%.

Keywords: Automated Valuation Model, Decision Tree, Boosting, Housing Markets