An Improved Analytical Center Machine for Classification

Abstract

AbstractAnalytical center machine (ACM) has remarkable generalization performance based on analytical center of version space and outperforms SVM. From the analysis of geometry of machine learning and principle of ACM, it is showed that some training patterns are redundant to the definition of version space. Redundant patterns push ACM classifier away from analytical center of the prime version space so that the generalization performance degrades, and slow down the classifier and reduce the efficiency of storage. Thus, an incremental algorithm is proposed to remove redundant patterns and embed into the frame of ACM that yields a redundancy-free, accurate analytical center machine for classification called RFA-ACM. Experiments with Heart, Thyroid, and Banana datasets demonstrate the validity of RFA-ACM.KeywordsVersion SpaceFeasible RegionGeneralization PerformanceTraining PatternIncremental AlgorithmThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.