Decomposition of price-forming factors based on statistical analysis of expert assessments of their significance

This study addresses the problem of mass appraisal of market value. In the context of deve¬loped markets, numerous studies have demonstrated the effectiveness of distinguishing, among all characteristics describing the object of valuation, those factors that are suitable for solving the classification task, while the remaining pricing factors are used to solve the regression task – namely, estimating the most probable price. Despite the demonstrated efficiency of this multi-model approach, there is no widely accepted methodology for separating descriptive features into classification and pricing categories. The aim of this study is to scientifically substantiate the effectiveness of decomposing features into classification and pricing groups through statistical analysis of expert assessments of feature importance. The hypothesis tested in this study suggests that features with high importance (as determined by mean, median, or mode) and high dispersion (i.e., a high coefficient of variation) should be treated as classification factors, whereas features with high importance and low dispersion should be treated as price-forming factors. Materials and Methods. Passenger vehicles are selected as the objects of valuation. The vehicle database used for deve-loping mass appraisal models contained three million records. To assess the influence of various vehicle parameters on price, a survey was conducted among 18 experts – professional market participants. The mean, median, and mode of expert assessments were used to rank the importance of the factors, while variance and coefficient of variation were applied for feature decomposition. A variety of mass appraisal models were developed using both classical regression methods and machine learning techniques (gradient boos¬ting, random forest, and artificial neural networks), with and without the separation of pricing factors. Results. The models that incorporated the separation of pricing factors demonstrated an average increase of 5–7 % in the coefficient of determination (R²) compared to models that used all features as a single set. Conclusion. The proposed approach to mass appraisal can be adapted to other types of valuation objects. The method based on statistical analysis of expert evaluations of feature importance is of a universal nature. This defines its practical significance, as appraisers and analysts in various domains often face substantial resource demands – time, computational, human, and financial – for constructing reliable mass appraisal models of market value.