The Research Of Bayesian Network And Cluster Algorithms Based On Weighted Feature

Along the fast development of the computer science and the continuous development of artificial intelligence, the computer is constantly used to simulate or replace a part of mental work for human, and provides great convenience to the people's life. Pattern recognition technique was born in the 1920's. With the further development of this technique and applied in many fields, pattern recognition technique promotes the progress of technology of the artificial, becomes a recent academic research pot. Classification, the supervised learning method, and clustering, the unsupervised learning method, are two main methods of pattern recognition.Classification is based on the training dataset and learns a classification model which can predicate the class label of an unknown sample. Clustering is a method based on some rules to cluster the samples of unknown class label, makes the samples in the same cluster more similar, and the samples in the different clusters less similar. This paper firstly proposes a simple learning method of a two-level Bayesian network structure. This method is not like the traditional methods which based on the searching algorithms and is a method based on the relationship between attributes. Through this learning method, it can get the relationships between class label and feature, feature and feature, and a Bayesian network structure, the network parameters, and finally can get a Bayesian network classifier. This method can avoid the searching process and also consider the relationships between attributes. The learning process is simple but the structure still fits for the data's true situation. Also, this paper considers the weight of different features when clustering. Besides, for different experiment results, this paper analyzed the result and proposed modified method and tested it.The experiments in two metabolomics mass spectrum datasets and five UCI public datasets proved the good performance the Bayesian network structure learning method which of this paper proposed. The experiment results in three UCI public datasets demonstrated the good performance of clustering based on feature weight, and also demonstrated the feasibility of the solution provided.