Impact And Evaluation Of Sample Selection In Incremental Decision Tree

With the advances in the technology of the database and network, people can easily get more and more data so that how to access and comprehend the data becomes an impending and challenging task. The methods in Machine learning can help us to understand and comprehend the knowledge hidden in the data which can guides and supervises our production practice. The major problems include the dynamically increasing data and the huge numbers of the data. The former can be resolved by the incremental learning method which updates the classifier by coordination algorithms when the new data come. The active learning can account for the latter through the active selection for samples which efficiently decrease the amount and complexity of the data in order to lessening the cost of classifier generation.In this paper, we study the active learning method based on the incremental decision tree through which combines the merits from the incremental learning and the active learning. The incremental decision tree is a effective incremental learning approach. It can update the decision tree with the adjustment algorithm which would destroy the minimum parts of the tree for keeping the structure stable. In terms of these works above, we provide a new sample selection algorithm, which implements the key part of the active learning, based on maximum numbers of the disagreement in classifications by taking account of the changing rules of the tree structure. At the same time, we introduce other two active learning methods based on the max entropy and most possibly wrong-prediction. The experiments show that the three algorithms obviously have efficient classification and decrease the needs of the samples for generating classifier. And it proves the validity of the active learning based on the incremental decision tree. In some cases, this method has better classification ability than the others.