基于最大共识的模型组合算法

Model combination algorithm based on consensus maximization

针对原有的随机森林算法没有区别各个单分类器之间的分类优势,对分类器的组合方案进行优化,提出一种基于最大共识的模型组合算法.该算法将分类器的经验误差和泛化误差融入到分类器的权重计算中,充分发挥了单分类器的个性与优势,强化分类效果好的单分类器的优势,弱化分类效果较差的单分类器的劣势.实验结果表明,基于最大共识模型组合算法能够提升组合分类器的分类性能,在提高分类精度的同时,也具有较强的泛化能力,这一改进对于提升同类型多模型组合算法的性能具有一定指导意义.

The combinatorial scheme was optimized and model combination algorithm based on the consensus maximization was proposed,aiming at the problem that the original random forest algorithm can not distinguish the classification advantage between each single classifier.The new algorithm integrated the empirical error and generalization error of the classifier into the classifier weight calculation,which makes each single classifier give full play to their personality and advantages.As a result,this method strengthened the advantage of good classifiers and weakened the disadvantage of poor classifiers.Experimental results show that this optimized algorithm can not only improve the performance of combinatorial classifiers,but also improve the classification accuracy and the generalization ability.This improvement is instructive to improve the performance of the same type of multi-model combination algorithm.