Forecasting House Resale Prices using Ensemble learning

Forecasting House sales prices is a critical issue for individual investors and government. prediction system has high accuracy and flexibility and plays an important role in decision making. However, previous studies have focused only on limited datasets, as well as nontime series or feature prediction. This paper proposes a blending model based on Python and its xgboost, DF21, and Geatpy packages to predict resale prices in Singapore. First, the high-cardinality categorical attributes are preprocessed by means of mean coding. Then, we propose a linear blending method that includes GA-HL-Xg-Boost, GARandom Forest (GA-RF), deep-Random Forest (DRF) and lightGBM, with Gini impurity to determine the importance of features. Finally, the result shows that it can obtain a mean absolute percentage error (MAPE) of 7.36% in a stationary overall trend of house prices trend. Our research can provide a powerful predictor of house resale prices in different economic environments.