چكيده به لاتين
Accurate house price predictions are essential for prospective homeowners, investors, appraisers, and insurers. However, the predictions of some studies are not accurate because they have not considered some factors, such as accessibility and economic attributes, that influence house prices. This study aims to predict house prices using structural, locational, accessibility, and economic attributes and finally explore the effect of the accessibility variable on the house price. The dataset contains 2,019,663 records of real estate transactions from 1975 to 2018 in the Washington metropolitan area, extracted from the Zillow website. In this study, the accessibility index is calculated using Distance, Cumulative Opportunities, and Gravity measures; gravity measure outperforms the others because it considers both land use and transportation components of accessibility. Then the economic attributes are used to predict the average house price of the study area each month by applying deep learning algorithms such as LSTM, GRU, and Simple RNN; a Simple RNN layer with 32 neurons is superior to the other models. After the structural and locational attributes are merged with the accessibility index and average house prices, machine learning algorithms, such as Linear Regression, Lasso, Ridge, Random Forest, GBM, LightGBM, XGBoost, Decision Tree, AdaBoost, Artificial Neural Network, and Stacked Generalization, are employed to predict the house prices. Finally, after evaluating the models, the Stacked Generalization (ANN + LightGBM) provides the best performance with R2 of 0.96 and RMSE of $23,290. This study also shows that when the accessibility index exceeds the threshold (80,003 in large buildings and 160,103 in small buildings), a higher accessibility index leads to lower housing prices due to noise pollution, decreased privacy, and increased supply responses.
