Enhancing Generalization Capability of Evaluation Model for Sound Quality by Regularization Method

Abstract

Regularization is known as a method that enhances the generalization capability to attain a high statistical accuracy with wide variety of data. Ridge regression, which is commonly used as the regression method with the regularization, is expected to be suitable for constructing sound quality evaluation models for the sounds emitted from industrial vehicles that produce various sounds, as the method does not reject explanatory variables. This study has revealed that Ridge regression has the higher capability to generalize the sound quality evaluation models on such industrial vehicles than Stepwise regression which rejects explanatory variables.