Comparative Study Of The Bayesian Classification And Improved Algorithm

An important aspect of human wisdom is its ability to know things outside. In order to master objective things, Things are composed by the degree of similar categories, the purpose of pattern recognition is to classify correctly about a specific things by using a computer. Classification is extensive research problems in then machine learning, pattern recognition and artificial intelligence related fields. In recent years, with new technologies are constantly emerging, the classification method also got new development.In view of the different classification problem, classification method varied. In many of the classification method, bayes classifier received great attention. In the model of bayesian classification, the model simulates conditional probability distribution of each category, then calculates posterior probability.based on bayesian theorem.But bayes classification instruments have stronger restrict which require between-class attributes are independent of each other, and classifier can not use between-class information effectively in training and learning process, however, between-class information for classification is important. Therefore, an improved algorithm of Bayes classifier combined with Fisher linear discriminant analysis is proposed. In this paper based on Fisher linear discriminant analysis this algorithm is the key to search optimal linear transformation matrix. Then the original sample are projected into the new sample space through the linear transformation matrix,and gain new samples. New sample in the new space can better separate, then these new samples are classifed by Bayes classifier. Experimental results show that combination of the classical bayes classifier and Fisher linear discriminant analysis method can get a better classification effect.In addition, this paper also research. further feature extraction and selection in pattern classification.At present,with rapid development of technology, information acquisition technology enhances unceasingly.the quantity of obtained data is quite large, and the dimensionality of data is also high. Feature selection and extraction can choose the most essential, the most relevant and the most effective characteristics in order to reduce data dimension and remove redundancy and unrelated features. Principal component analysis and K-L transformation is the basic feature extraction method. In order to realize the data dimension reduction,we can use standard orthogonal transformation through the original data set.Through the experimental simulation, experimental results show that the feature extraction and selection helps improve the classification effect and improve classification accuracy.