Selective Ensemble Learning Research Based On Q Statistic

Machine-learning technology has been successfully applied in all aspects of social life. Such as handwritten Chinese character recognition, face recognition, network intrusion detection. As machine learning has made great achievements in the applications,many researchers have carried out the study on machine learning. Ensemble learning in the machine learning is an important research direction. The studies have shown that ensemble learning can significantly improve the generalization ability of machine learning systems. As ensemble learning needs multiple learners, it is greater computational complexity. However, with the increase in individual learner, the diversity between learners is smaller, and ensemble learning accuracy is worse. The Nanjing University Professor Zhi-Hua Zhou proposed the concept of selective ensemble learning. It selects some learners which are a part of the original learners. Experiments show that the effect is better than the original ensemble.The diversity in the ensemble learning is an important concept. The key of a good ensemble learning algorithm is able to generate the diversity of individual learners. So far, researchers have been raised 10 kinds of different diversity. As the Q statistic is better in experiments, so this paper uses Q statistic to measure the diversity between the two learners and proposes a new selective ensemble algorithm basing Q statistic.As the Weka is today's most popular machine learning platform, it offers a variety of machine learning algorithms, as well as the assessment tools for experimental results. Therefore, the experiments are all based on Weka platform. Meanwhile, in order to use the powerful features of the Weka platform, the algorithm is integrated into the Weka platform. Finally, the paper also shows the problems in the algorithm description and outlook, it will facilitate further research. Generally speaking, this paper contains several aspects:(1) It shows an overview of selective ensemble learning in international research status, background and significance. It includes the origin, definition, the main technical, and diversity of the selective ensemble learning. Also it introduces the open-source machine learning platform for Weka.(2) From the view of the diversity, it puts forward a new selective ensemble learning. This method uses the popular Q-statistic. In order to prove the validity of the algorithm, it uses decision tree as the base learner. The UCI data sets are used to do experiments, and the experiments show that the algorithm can not only reduce the number of learners, but also can improve the generalization ability of ensemble learning.(3) To improve the convenient and efficient of the algorithm, the QSE algorithm is integrated into the Weka platform. This is not only conducive to the promotion of this algorithm, but also more machine learning researchers can communicate and learn each other.