Research On Label Membership-and Slowness Principle-Based Ordinal Discriminant Regression With Applications

Ordinal regression is a special supervised learning paradigm in pattern recognition,the objective is to establish a regressor based on a given set of input and ordered discrete output,so that it can predict the sample class of ordered discrete test.In practical application,this kind of standard discrete and orderly scene exists widely;the scalar value is usually based on people's preference.This label not only reflects the level difference between ordered labels,but also reflects the connection between different label samples.Therefore differencing from the simple regression and classification learning,OR has dual characteristics of classification and regression.Compared to the ordinary regression method,the ordinal regression has better performance because it uses the priori information that class label has order.In recent years,Ordinal regression has received more and more attention and research because of the extensive application of OR in face recognition,credit rating,credit rating,brain computer interface and age estimation field.However,there are still some deficiencies existing in OR methods in terms of the use of prior information.To this end,this paper has carried on the research in the following several aspects:In terms of the fact that existing OR method has problem in using label membership information between ordered label,we designed a kind of quantitative representation for the membership information,and then modeling for ordered learning through combining the quantitative representation with classical and effective regression method KDLOR(Kernel Discriminant Learning for Ordinal Regression).The method been constructed called Linear Discriminant Learning for Ordinal Regression using Label Membership,LM-LDLOR.Furthermore,we derive the kernel discriminant learning for ordinal regression using label membership(LM-KDLOR)in order to solve the nonlinear situation.At last,the effectiveness of the proposed strategy is verified by comparing experiments on 8 standard ordinal regression data sets.The slowness principle is a learning principles constructed on the visual characteristics of the human visual system,and it has wide applications in pattern recognition.However,as far as we know,slowness principle has not been studied by combining it with ordinal regression.Inspired by this,this paper forms SP-DLOR(Slowness principle based discriminant learning for ordinal regression)through using nearest neighbor method to create the slowness within-class scatter matrix,also assurance class label in order by ordered constraint and achieve orderly learning by according to linear discriminant criterion.Finally,the comparison experiments on 8 standard ordinal regression data sets and FG-NET face dataset show the superiority of the proposed algorithm in regression and classification performance.