Predicting Model For Urban Residential Housing Prices

Abstract

The need to determine house prices beforehand is an important element in making a decision on whether to purchase a house or not. The commonly used price forecasting models are single predictor models but are prone to over-fitting and low accuracy levels emanating from their inability to handle noisy data. We propose a regression based ensemble learning model that incorporates multiple predicting models while using Root Mean Squared Error (RMSE) and R-Squared error to measure model performance. This entails leveraging on extreme gradient boosting (XGBoost), Random forest and Light gradient boosting (lightGBM) algorithms to form base models. Stacking of the models was also used before generating the final model using weighted voting. The dataset used is the Ames housing dataset readily available on Kaggle platform. The results of the study reinforce further that ensemble learning method greatly helps improve accuracy of the model as compared to single prediction models