A New Hidden Markov Model with Application to Classification
Changshou Deng, Pie Zheng · 2006
A new hidden Markov model was proposed by using the framework of a Markov chain to deal with classification. Based on this framework, a new estimation method for the transition probabilities among the hidden states was discussed, which avoids the local maximum led by the learning method of the traditional hidden Markov model. Using the stationary distribution of the hidden states, a classifier was proposed with observations being easily classified. Numerical examples were given to demonstrate the initial use of the model by using the standard data set. The result shows the effectiveness of the model-based classifier. The new model-based classification method can be widely used in statistical time series analysis such as speech recognition and handwritten characters recognition