Optimization Method Research And Application Of Multiple Classifiers Ensemble Based On Diversity Measure

Classification is one of the most important basic activities of human and played an important role in our daily life,social activities and work.With the rapid development of data mining and pattern recognition,Researchers apply machine learning and pattern recognition to analyze and process data,finding inner special relationship between data.It is of great guiding significance and application value to solve the classification problem in practical application,promoting the rapid development of classifier technology.Classifier has experienced the development process from single classifier to multiple classifier ensemble.In 1997,T.G Dietterich,the international machine learning field authority,pointed out that ensemble learning,symbolic learning,statistical learning and reinforement learning make up four major research direction of machine learning.Ensemble learning which is listed as research direction,is a new paradigm of machine learning.But now,how to train the basic classifier with higher accuracy and greater diversity?How to realize more effectively ensemble learning?These are the unsolved problem in ensemble leaning area to improve the generalization ability of ensemble learning system.Above background,this paper carried out the related research focusing on classifier ensemble optimization method and application based on diversity measure.The main research work and achievements is as follow:(1)A diversity measure method,based on distance entropy,is proposed.In consideration of combing distance with information entropy,distance entropy which is kinds of evaluation indexes of basic classifier relative baseline data is utilized to measure overall diversity.(2)A selective ensemble method of hybrid heterogeneous classifier,based on rotation forest transformation,is proposed.This method constructs training subsets based on feature partition and multiple heterogeneous classifiers by the rotation forest model.And it make use of distance entropy to measure diversity.This paper determines the weights of data fusion by confidence degree of classification results,which is accuracy and diversity of basic classifier as a selective integration condition.(3)This paper applied the classifier ensemble system to detect human action based on heart rate change.