Hybrid soft computing methods for improving real estate price forecasting

Real estate has a very important place for countries and the world economy. The fact that real estate has a heterogeneous structure, that is, the diversity of the properties that make up each property itself, shows that the methods to be selected for determining the sale and rent value of the real estate are important. For real estate price estimation, it can be said that non-traditional methods are more successful than traditional ones. In recent years, metaheuristic approaches for real estate valuation have increased and researches on this subject are continuing. The superiority of metaheuristics in solving non-linear and complex problems provides an important advantage for real estate valuation. In this study, a hybrid approach has been developed for real estate valuation by using artificial neural networks and feature selection methods. The grid search method was used for the most suitable model parameters before the analysis with the artificial neural network. It is aimed to increase the model success by optimizing the parameters that can be selected for the model with grid search method. In addition, in order to minimize possible errors and prevent overfitting during training, the data was divided into appropriate training sets by cross validation technique. The selected data set for this study contains housing data of various districts of Istanbul province, which has a significant value for Turkey. Considering that data preprocessing is important for the success of the model, data preprocessing steps were performed for each district data. After the data preprocessing process, firstly feature selection method, then grid search method with cross validation and finally artificial neural network method with cross validation were applied step by step. The advantage and success of the artificial neural network method, which is one of the metaheuristic approaches, is supported by this study for real estate price estimation involving many variables. With this model developed for real estate appraisal, it is aimed to examine the properties that affect real estate prices and to contribute to valuation methods in addition to finding realistic price estimation. In addition, it is an important advantage that it is possible to follow the price changes in the real estate market which has a heterogeneous structure with this research, and it is thought that it will provide benefit for the subsequent real estate appraisal studies.