Building of Semi-Quantile Regression Models

Abstract

The article describes the mathematical background of building of semi-parametric quantile regression models (LMS-models). The basic characteristics of statistical distributions are considered and the algorithms of their transformation to normality (box-cox transformation, Manly exponential transformation, modular transformation) are given and their modification is proposed. The criteria for selecting optimal models are defined. The developed integrated algorithm of LMS - model construction based on the transformation of the primary measurements to normality, including stages of the model initialization, the model selection and the model configuration is presented. The scheme of a process of the model configuration with the use of a modified chain graph is introduced. The reference chart of body mass, which can be used to monitor obesity in the pediatric population are built with the application of the proposed LMS-method in different quantile modes.