Bulk Loading Hierarchical Mixture Models for Efficient Stream Classification

Abstract

AbstractThe ever growing presence of data streams led to a large number of proposed algorithms for stream data analysis and especially stream classification over the last years. Anytime algorithms can deliver a result after any point in time and are therefore the natural choice for data streams with varying time allowances between two items. Recently it has been shown that anytime classifiers outperform traditional approaches also on constant streams. Therefore, increasing the anytime classification accuracy yields better performance on both varying and constant data streams. In this paper we propose three novel approaches that improve anytime Bayesian classification by bulk loading hierarchical mixture models. In experimental evaluation against four existing techniques we show that our best approach outperforms all competitors and yields significant improvement over previous results in term of anytime classification accuracy.KeywordsMixture ModelData StreamGaussian Mixture ModelMixture DensityHilbert CurveThese 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.