A comparative analysis of attribute reduction algorithms applied to wet-blue leather defects classification.

摘要

This paper presents an attribute reduction comparative study on four linear discriminant analysis techniques: FisherFace, CLDA, DLDA and YLDA. The attribute reduction has been applied to the problem of leather defect c1assification using four different c1assifiers: C4.5, KNN, Naive Bayes and Support Veetor Machines. Results and analyses on the performance of correct c1assification rates as the number of attributes were reduced are reported.