Dynamically Ensemble Rough Set Reducts

Ensemble Learning and rough set theory are hot topics in the field of machine learning and artificial intelligence in the world of today. Many studies on the ensemble indicate that combining classifiers has a good result of learning. How to build better sub-classifiers and how to select sub-classifiers to ensemble are key factors in ensemble learning. Rough set is an important learning tool with solid mathematical foundation, and has been widely applied in pattern recognition, data mining and many other fields. Based on the classical rough set theory, people have put forward many new methods, which make the rough set method has better performance as a learning tool.We focus on the integration of the advantages of rough sets and ensemble to enhance learning effect. In this paper, we firstly introduce the principle of rough set and how to get the roughset reducts and rules, and then introduce some measure of rules and knowledge of ensemble. Finally we propose a trust-based dynamic integration of rough set reduction approach. Using the method of rough set to generate some sub-classifiers which have certain difference among them and preserve classification accuracy, at same time using the dynamic integration strategy, we try to resolve the problem "when at least one classifier can correctly classify the testing samples how can we improve testing accuracy". The results of the experiments show that our method is effective. We also provide analysis for some interesting phenomena in our experiments.