Abstract. The purpose of this article is to examine the prediction accuracy of the Random Forests, a machine learning method, when it is applied for residential mass appraisals in the city of Nicosia, Cyprus. The analysis is performed using transaction sales data from the Cyprus Department of Lands and Surveys, the Consumer Price Index of Cyprus from the Cyprus Statistical Service and the Central Bank of Cyprus' Residential Index (Price index for apartments). The Consumer Price Index and the price index for apartments record quarterly price changes, while the dependent variables for the computational models were the Declared and the Accepted Prices that were conditional on observed values of a variety of independent variables. The Random Forests method exhibited enhanced prediction accuracy, especially for the models that comprised of a sufficient number of independent variables, indicating the method as prominent, although it has not yet been utilized adequately for mass appraisals.

Property tax in Greece is levied since 1985 not on Market Values but on the "objective value" of the properties as it is defined by the Ministry of Economics. It forms a non-flexible system, with market-irrelevant and unrealistic values, inducing land-policy practices and potential political cost to each periodical update. Furthermore, instead of adjusting taxation levels to the current economic reality, the real estate market is experiencing further burdening through approximately 40 different property taxes and levies, leading to further shrinking and depreciation. The authors believe that a fairer taxation system could significantly assist the property sector in Greece. Thus, through this paper and by studying and analyzing best practices from other countries, they propose models that can be applied with the use of existing data in Greece. This work aims to identify the critical parameters that affecting property values in Thessaloniki to create a Market Value forecasting tool for a fairer taxation system, to highlight the importance of a GIS system for this purpose and to compare the results of MRA with the use of SPSS with those of GWR in ArcGIS environment. For the purposes of this study, the Municipality of Thessaloniki was chosen due to its very well organized portal with significant and well organized geographical data and because authors manage to access some data from the Central Bank of Greece, regarding property valuations.

A recent study of property valuation literature indicated that the vast majority of researchers and academics in the field of real estate are focusing on Mass Appraisals rather than on the further development of the existing valuation methods. Researchers are using a variety of mathematical models used within the field of Machine Learning, which are applied to real estate valuations with high accuracy. On the other hand, it appears that professional valuers do not use these sophisticated models during daily practice, rather they operate using the traditional five methods. The Department of Lands and Surveys in Cyprus recently published the property values (General Valuation) for taxation purposes which were calculated by applying a hybrid model based on the Cost approach with the use of regression analysis in order to quantify the specific parameters of each property. In this paper, the authors propose a number of algorithms based on Artificial Intelligence and Machine Learning approaches that improve the accuracy of these results significantly. The aim of this work is to investigate the capabilities of such models and how they can be used for the mass appraisal of properties, to highlight the importance of sensitivity analysis in such models and also to increase the transparency so that automated valuation models (AVM) can be used for the day-to-day work of the valuer.