Chapter 8 - Search of regularities in data: optimality, validity, and interpretability

Abstract

The chapter discusses a method for verifying regression models based on the Occam's razor principle. A variant of the formulation of the Occam's razor principle is used to make choice between simple and complex regression models. A formulation of Occam's razor principle is used to choose between simple and more complex regression models. Randomized permutation test is used to test null hypothesis that the simple model is sufficient to describe relationship existing in data. A criterion statistics is calculated using more complex regression model. The chapter discusses in detail examples of the successful application of the Occam's razor principle for the verification of piecewise linear regression models and models based on optimal partitions of joint ranges of admissible values for pairs of features.