Research On Multi-Dimension Bayesian Network Classifiers Based On Feature Selection

Multi-dimensional data classification is one of the important research directions in data analysis. Because of the increase of data attributes, data often contain many confusion variables and the feature selection is introduced to the learning of classification. With the further study, the data obtained are always missing for various reasons. Algorithms for multidimensional Bayesian classification are proposed to address incomplete data, but these methods have certain limitations or the pre-assumption of data missing at random. In this paper, we combine the two feature selection algorithms with the multi-dimensional extended Robust Bayesian classifier to classify the classification problems under non-random missing data.Firstly, Electromagnetism-like Mechanism as a method of feature selection is introduced to select attributes from data by imitating the interaction of electric charge in electromagnetic. Then some redundant and non-related variables are removed for feature reduction. The computational complexity of the classification model is reduced and the classification efficiency is improved. The extended multi-dimensional Robust Bayesian classifiers is proposed. And a mathematical model is established for the multi-dimension classification problem according to the obtained attribute subset under the non-random missing data. Instead of directly estimating the specific value of the posterior probability, the method is used to approximate the posterior probability, and the approximate value is used as the evaluation criteria to classification forecasting. In order to test the model's advantages and disadvantages, the simulation experiments are carried out on the three kinds of common multi dimension data sets. This method proposed in this paper is compared with other algorithms with the index of classification accuracy. The results show that the method proposed in this paper has good classification performance.Then ant colony algorithm, a very popular intelligent algorithm, is applied to feature selection. This approach is a meta-heuristic algorithm which was inspired by the foraging of real ants. Individual artificial ant is corresponding to a solution and each feature variable represents a node in the solution to seek the shortest path. Ants traverse all nodes by applying certain rules to determine whether each node is selected or not. The final solution is that the attribute subset is determined. In this paper, an improved ant colony optimization algorithm is used to select the features of the problem, and then to establish a multi-dimensional classification model for classification. Then the class variables of multi-dimension Bayesian classifiers are defined as a composite variable using the attribute subset obtained previously. The problem can be turned to one-dimensional problem. The simulation results show that the method has good classification results compared with other algorithms.