Neurons Classification Based On The Bavesian Classifier

Naive Bayesian classifier is a simple and efficient classifier. It can obtain the sameclassification accuracy of some complex classification in many case. However, because theattribute independence assumption does not hold in the real problem, focusing on how to relaxthe assumption of independence, and also has a good classification result is a key to improvethe naive Bayesian classifier. The main content and innovation are as follows:1. This paper presents a naive Bayesian classifier attribute selection algorithm that basedon partial least squares method. Through establishing the partial least squares regressionequation between the conditions of property, then get a regression coefficient matrix.Normalized and sum it, got the relevance of the property. The greater the correlation is, theworse this property is.2. Use partial least squares correlation analysis to select the attribute can get thecorrelation between each attribute, but the classification of each property’s good or poor can’tbe judged. This paper proposed an extraction method based on the each attributes of theproperty value of interval relations. In the Naive Bayes model, the value range which in thedifferent categories of the same property is different. If it is too close to the property that can’tdistinguish between these two categories. Then count the number of categories which theproperty can’t distinguish it. The fewer the number is, the better description of the propertyclassification is.3. This paper presents a probability-based weighted Naive Bayes classification algorithm.Use naive Bayesian classifier to class the each attribute, then get the probability of correctclassification of the property, Finally, the probability as the corresponding weights were addedto the conditions on the properties to get the weighted Naive Bayes classifier.4. Based on the above three methods, use the data of the neurons as the example, selectedthe properties, and weighted the classifier, then get the better satisfactory results. Thecross-validation classification accuracy rate is increased by16%. Using Weka data sets withthe same attribute classification, weighted classifiers is better than not weighted...