Abstract:

Hidden Markov Model (HMM), a statistical machine learning techniques, while well proven remarkable success in many fields in the past decade; such as speech recognition, handwriting recognition, bioinformatics, and information extraction, are just beginning to be applied to Supervised learning problems. We propose a new method to improve the classification performance, which is beyond methods take statistical parametric in their estimated approach. The constructed model is based on estimation techniques of HMM, where the classification is formed by only assigning probabilities expressed by a linear function of observed characteristics. Classify with hidden markov model such as statistical methods is equivalent to determine a discriminative model, which endows our method with a solid probabilistic interpretation. We also show that use of model in real data to set model parameters provides a significant improvement in classification accuracy and simplify the use of data. Experimental results show that the new method has strong performance in classification purity for various datasets, especially for large-scale manifold data.