摘要： 根据基于类的特征向量方法的原理，提出了一种基于投票的叠加泛化方法，对0-层分类器的预测结果“投而不决”，由1-层分类算法来归纳投票情况与正确类之间的关系。实验表明，该方法在具有明显类分布倾斜的多类数据集上有较理想的泛化效果。

关键词: Stacking；分类器组合；投票

Abstract: This paper describes a new strategy for forming class vector, which employs voting mechanism to “vote but not make” the final predictions of all the level-0 classifiers. In this approach, the relations between votes and the true class are induced by classification algorithm of level-1. Experimental result shows that it achieves good performance on those datasets with highly skewed class distributions

Key words: Stacking; Classifiers combination; Vote