Evaluating machine learning algorithms for predicting house prices in Saudi Arabia

Abstract

With real estate projects increasingly growing, it has a critical economic role in countries and a significant effect on economic and investment domains. There is limitation of researches have contributed the inclusive evaluation of machine learning algorithms in real estate price prediction in this field. Machine learning algorithms could assist in forecasting real estate prices; thus, this paper presents a comparative assessment of machine learning algorithms to predict housing prices in Saudi Arabia. A real estate transaction datasets is used to evaluate the accuracy of machine learning algorithms. The datasets contains 13 columns and 59,846 instances (records). The results considerably vary in performance across the evaluated algorithms: random forest (RF), decision tree (DT) and linear regression (LR). After extensive examination the three proposed algorithm, Random Forest obtained R2-score 0.65, the decision tree obtained R2- score 0.30 and linear regression obtained R2-score 0.06. Random Forest performed better than the decision tree or linear regression methods.