Research Of Multiple Classifiers Ensemble Learning Method Based On Rough Set

Ensemble learning is current research hotspot in the field of machine learning. Through the combination of multiple classifiers to solve the same problem, it can obtain better performance than using just a single classifier. Because Ensemble learning has a good generalization performance, it has been applied in many fields, such as face recognition, speech recognition, computer vision and so on. Rough set theory, established by Polish scientist Z. Pawlak in 1982, is one of the main analysis of uncertainty theory. It provides a complete analyzing and processing for the classification of the data. At present, because of its unique advantages, rough set theory has been widely used in pattern recognition, machine learning, knowledge discovery, data mining and other fields.On the basis of comprehensive introduction to multiple classifiers ensemble learning and rough set theory both at home and abroad, this thesis combines the characteristics of both. In this thesis, the rough set theory is introduced into the ensemble of multiple classifiers, launched an ensemble learning method based on rough set research. The main contents are as follows:First of all, in order to combine the rough sets theory and ensemble learning more effectively, and improve the classification performance of multiple classifier ensemble, a multiple classifier ensemble learning algorithm which combines rough set with ensemble learning is proposed in this paper. Based on rough set theory, samples are divided into positive region and boundary region, which is useful for training data sampling. During the sampling process, make sure that all samples within the boundaries are included in each sampling data set. Experiment results on UCI data set indicate that the proposed algorithm can get better performance as for precision and recall when compared to traditional methods.Secondly, in view of the dynamic data ensemble learning study, combined with rough set theory and incremental learning, this thesis prpposes an incremental ensemble method based on rough set. This method which will be in the process of the rough set theory is introduced into the incremental classification. It not only can make the classifier after ensemble to adapt to the constantly changing data, incremental learning effectively, but also reduce the training time and consumption of storage resources. On UCI data sets of the experimental results show that comparing to some incremental and non incremental ensemble method, the method can improve the effectiveness of classification.