A Hybrid Machine Learning Model Architecture with Clustering Analysis and Stacking Ensemble for Real Estate Price Prediction

Population growth, rapid developments in technology, increase in living standards, changes in the household structure and economic structure of societies, and the increase in urbanization at very high rates, as well as the increase in the demand for renting or purchasing real estate, have both expanded the real estate market and made it more active. This intense activity in the real estate markets also accelerates real estate price prediction studies in direct proportion. The aim of this study is to present a model architecture that can achieve high accuracy in predicting the current market value of real estates by using a hybrid approach, through clustering models as a preliminary approach, in order to achieve higher homogeneity with stacking ensemble using multiple machine learning methods. In order to obtain more homogeneous submarkets, the collected data set was first grouped according to the number of rooms and then each group was divided into clusters by cluster analysis. In this way, more homogeneous submarkets were obtained and predict accuracy was improved. Then, the training process was carried out for 13 different weak learners using fivefold cross-validation for each determined sub-market. Feature selection and parameter optimization were performed separately for each weak learner. Then, the predictions obtained according to the feature and parameter set that gave the best results were used to train the meta-learner. As a result of this entire process, the final prediction was created with the meta learner that gave the least error rate. As the findings show, high predicting performance at international standards has been demonstrated even in a period of high price fluctuations for many and various sub-markets of real estate.