A geometric framework for multiclass ensemble classifiers

Abstract Ensemble classifiers have been investigated by many in the artificial intelligence and machine learning community. Majority voting weighted majority are two commonly used combination schemes ensemble learning. However, understanding of them is incomplete at best, with some properties even misunderstood. In this paper, we present a group these formally under geometric framework. Two key factors, every component base classifier’s performance dissimilarity between each pair evaluated same metric—the Euclidean distance. Consequently, ensembling becomes deterministic problem an can be calculated directly formula. We prove several theorems interest explain their implications for ensembles. particular, compare contrast effect number on types schemes. Some important both discussed. And method to calculate optimal weights presented. Empirical investigation conducted verify theoretical results. believe that results from paper very useful us understand fundamental principles general. The also helpful investigate issues classifiers, such as prediction, diversity, pruning, others.