The original classification algorithm for the improvement of regression models for the purpose of taxation

Abstract

Quality of a practical regression appraisal in some cases can be improved by deЇnition and subsequent division of the market sample into two latent classes of "cheap" and "expensive" objects and by constructing models for corresponding classes. Before the calculation of the objects' prices each of them has to be ascribed to one of two classes deЇned. Several methods of classiЇcation were analyzed and original algorithm, named KL, was developed, which possesses a few important advantages in comparison with recognized kNN and C4.5 algorithms. Described approach has proven e®ective on real data used in mass appraisal. Essentially, low error rate of classiЇcation determines high quality and fairness of regression appraisal for the purpose of taxation.