Assessment of the Hybrid Deep Learning Models and Hedonic Pricing Model for House Price Prediction

Modeling the pricing process or estimating a property’s fair market value in an objective manner is highly complicated. Several variables, including geographical and temporal ones, can heighten this intricacy. Researchers and real estate appraisers have been attempting to simulate the process for ages. Prior until recently, real estate appraisal and data evaluation were both improved by computer-aided valuation systems. However, they are not very efficient, accurate, or transparent and require continuous improvements. In this study, we investigate how hybrid deep learning (DL) algorithms can improve economic activity by making better predictions of house prices. The research employs the DL algorithms and the hedonic regression modeling for the house price prediction. DL and hedonic pricing models were trained using the Ames housing dataset containing 82 variables and 2,930 samples. The research contributes to the feasibility of employing hybrid DL algorithms to predict home values. Finally, the prediction model’s goodness of fit is reported, with CNN+LSTM+FCL (the proposed hybrid DL model) achieving the highest R-squared score of 96.15%, while the hedonic regression model acquired only 52.03%.