Usage of machine learning algorithms in housing value estimation: Ankara, Gölbaşı application

The need for determining the value of real estate in our country and in the world is increasing day by day. Real estate appraisal studies, which are an important part of the national economy, are encountered in many applications such as taxation, privatization and expropriation. There are many studies in the literature to find the estimation approach in the valuation of immovable properties. Various criteria were used in these studies according to their application areas. Their success levels vary according to the criteria and methods used. In this study; 95 samples of housing were collected from 5 neighborhoods determined from Gölbaşı district of Ankara province. As a result of the literature research carried out to determine the criteria expressing these samples in the appraisal, the main data set was created with 42 criteria that are affecting the value of the real estate. The main data set prepared for valuation; different data sets were prepared after criteria reduction, data processing, normalization and weighting steps. In this study, Support Vector Machines (SVM), Support Vector Regression (SVR), Multiple Linear Regression (MLR) methods were applied by using open source Orange Canvas program and RMSE, MAE, R2 performance results obtained from the application were compared. In addition, using the data sets prepared, the same methods were used to determine the weights of the criteria affecting the value of the property. When the success of the models were compared, more successful results were obtained in the data sets prepared by weighting the criteria. Among the applied methods, it was observed that the performance result values obtained by the SVM method were more successful than the other methods and it was observed that the SVR method also reached very close results. In the weight analysis of the criteria that affect the value of the immovable, it was seen that the weight of the criteria varied in each data set and each method applied. As a result, the usability of SVM, SVR and MLR methods used in this study has been demonstrated that is an alternative valuation method against classical methods used in real estate appraisal. It has also been found that SVM can be used to create a collective valuation system. Keywords: Machine Learning, Multi Lineer Regression, Real estate valuation, Real Estate Criteria, Support Vector Machine, Support Vector Regression, Valuation.