Multi-Label Classi Ication Based On Label Relations And Its Application In Diagnosis For Parkinson

In Chinese Scale for Parkinson(CSP),judging Chinese Syndromes(CSs)in term of symptoms in itself a typical technology of multi-label classification.The research idea of this article treats symptoms in CSP as feature attributes and CSs in CSP as labels,and the conclusion associate feature attributes with labels is judged from algorithms of multi-label classification(MLC).For Parkinson dataset,the main works of this article are mentioned as follows:1)A new MLC algorithm called Prim based Classifier Chains(ECC)based on Classifier Chains model has been proposed to solve the side effect of randomness that exists in CC model on accuracy of classifier and can optimize the order Chains.Because of it,the relationship among CSs can be discussed.According to the principle of it,this algorithm,lower is the information entropy,lower is the uncertainty,and higher is the probability of accurate prediction of labels so that the label which has a minimum value of information entropy should be the head of the order Chains which is the start node that is used for constructing the minimum spanning tree(MST)As a result,the optimized order Chains traversed from MST will be applied to CC.2)The mechanism of Josephus Ring(JR)has been introduced to generate a new algorithm called Prim and Josephus based Classifier Chainss(JCC)based on ECC.The order Chains based on label correlation and traversed from MST,however,is not a order globally and still exists randomness which should be reduced by JR.3)The mechanism of penalty has been introduced to be a counting method on the basis of JCC,producing a new algorithm of multi-label classification called Penalty based Classifier Chainss(PNCC),which can further reduce the randomness that is reduced finitely from ECC to JCC and thus produce a label order Chains that will be applied to CC.And Experiments show that ECC and JCC possess highly competitive performance on Parkinson dataset and pubic dataset.