A sequence classification methodology is proposed based on a conjecture that neighborhood' s similarity results in sequence' s similarity.The constrained Hidden Markov Model(HMM) space defined by sample is transformed to unconstrained HMM space.The neighborhood information is extracted at the standard HMM,and is imported to the SVM.Experimental results show that compared with other classical sequence classification methods,the proposed methodology can indeed greatly improve accuracy or speed.Meanwhile,the results also validate the original conjecture.
sequence classificationunconstrained spaceneighborhood informationHidden Markov Model (HMM)Support Vector Machine (SVM)