| Classification is one of the most important tasks in data mining. In the commonly used classification methods, the analysis of Mahalanobis-Taguchi System is based on data while not need to make assumptions about data distribution, and can selects feature variables effectively, so it can reduce dimensionality in the true sense. Because of these characteristics, it now widely used in many fields. This article mainly studies two-class classification problem. The main work includes the following points.(1) Research on Mahalanobis-Taguchi System based on particle swarm algorithm for two-class classificationTo select the feature variables, the traditional Mahalanobis-Taguchi System often use the orthogonal table and signal to noise ratio method, and use the loss function method to determine the classification’s threshold. But the some researchers indicated that the method of orthogonal table and signal to noise ratio was not the best. In addition, the loss function had strong subjectivity to determine the threshold. This paper researches on Mahalanobis-Taguchi System based on the particle swarm optimization (based on balanced data), using particle swarm optimization algorithm to choose characteristic variables and determinate the threshold.(2) Research on Mahalanobis-Taguchi System based on Bagging for imbalanced data classificationIn real life, there are exist many imbalanced data classification problems, imbalanced data classification problem refers to the classification of samples have significant difference, while at the same time people often pay close attention to a few sample class. The traditional Mahalanobis-Taguchi System of clarifying unbalanced data, the classifier favors seriously to the majority class. In order to improve the accuracy of imbalanced data classification, we put forward applying Mahalanobis-Taguchi System combined with optimization and integration to reduce the classifier’s tendency to the majority class and the dimensions of data. |
PDF Full Text Request