Reducing classification times for email spam using incremental multiple instance classifiers

Abstract

Combating spam emails is both costly and time consuming. This paper presents a spam classification algorithm that utilizes both majority voting and multiple instance approaches to determine the resulting classification type. By utilizing multiple sub-classifiers, the classifier can be updated by replacing an individual sub-classifier. Furthermore, each sub-classifier represents a small fraction of a typical classifier, so it can be trained in less time with less data as well. The TREC 2007 spam corpus was used to conduct the experiments. © 2011 Springer-Verlag.