Methodological aspects of practical regression appraisal: improvement of model's estimation properties by means of quantitative attribute values coding

Preceding part of this paper posed important problems concerning practical regression appraisal. The present part suggests heuristic procedures to cope with the two problems. The first one is connected with the choice of preliminary nonlinear transformation of quantitative attributes. The transformation method, which uses the notion “proximity of values” and allows to considerably improve the model quality is proposed. The second problem concerns the development of models with enhanced prognostic features, examined on a test sample. Several interconnected empirical procedures are suggested on training and control samples for consecutive selection of significant regressors to develop models with requested properties. Effectiveness of all described methods is demonstrated on the practical example of mass appraisal of land parcels in the Kaluga region of Russian Federation.