Research Of The Continuous Attributes Bayesian Classifier Based On Spatial Distribution Information

Bayesian classifier is a kind of Bayesian network for classification, it has solid theoretical foundation and good classification performance. Thus it has been widely concerned. When dealing with continuous attributes classification problems using Bayesian classifier, the first step is a discretization, which will increase the consumption of calculation and the process will also lost part of classified information, especially the training sample size is small, these factors affect the performance of the Bayesian classifier for continuous attributes sample. In order to promote the performance, this paper use the method of estimating attributes' probability density to construct a continuous attribute Bayesian classifier, which can directly deal with the continuous attributes and get more useful information. Considering the continuous attributes distribute are easy to obtain spatial probability distribution, and spatial probability distribution is often used to mining latent variables or dependent extensions, which is the best separability measure tool. This paper use the spatial distribution information of attribute variables that including the error rate and the inter class distance to improve and construct a new continuous attribute Bayesian classification model.The derived classifier of Naive Bayesian try to relax the independence assumption for get more loss information. The weighted attribute by quantifying the different attributes' contribution to the decision to improve the performance of the classifier and it also maintain good computational efficiency. This paper uses error rate method to construct a weighted classifier. In order to improve the imbalanced data classification accuracy of the minority class, this paper improves the local prior probability calculation by distance separability measure method and search the optimal intermediate parameters by setting the minority class preference to construction a new Bayesian classifier. The new classifier uses a two-level decision to enhance the classification performance of the minority class.The main content of this paper is to improve the classification performance of continuous attributes Bayesian classifier, and the main research work as follow.Using weighted attribute improved the naive Bayesian classifier with continuous attributes, discusses influence of weighted to the Bayesian classifier, introduces the classification error rate calculation method. Settings weight by the error rate and construct weighted naive Bayesian classifier. Finally, verifying the new classifier performance by the contrast experimental and analysis experimental results.This paper propose a specially continuous attribute Bayesian classifier for adapted to the imbalanced data sets. Firstly, ranking attributes by separability measure based on distance, use the joint probability density estimate the poor separability attribute. Considering the efficiency of the algorithm, searching the optimal multiple window width with Monte Carlo sampling method. In this paper, the algorithm is presented in detail process. Finally, evaluation of the classifiers and given analyzed.