基于最大共识的模型组合算法

发布时间：2018-04-27 12:12

本文选题：随机森林 + 最大共识　；参考：《浙江大学学报(工学版)》2017年02期

【摘要】：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.
[Abstract]:Since the original stochastic forest algorithm does not distinguish the advantages of each single classifier, the combinatorial scheme of the classifier is optimized, and a model combination algorithm based on maximum consensus is proposed. In this algorithm, the experiential error and generalization error of classifier are integrated into the weight calculation of classifier, and the individual character and advantage of single classifier are brought into full play, and the advantage of single classifier with good classification effect is strengthened. Weakening the disadvantage of single classifier with poor classification effect. The experimental results show that the combination algorithm based on the maximum consensus model can improve the classification performance of the combined classifier and improve the classification accuracy as well as the generalization ability. This improvement has certain guiding significance for improving the performance of the same type multi-model combination algorithm.
【作者单位】：吉林大学计算机科学与技术学院;东北师范大学计算机科学与信息技术学院;
【基金】：国家自然科学基金资助项目(61272209)
【分类号】：TP301.6

【相似文献】