A Novel Framework for Spam Hunting by Tracking Concept Drift

Abstract

In mid seventies a new method of exchanging messages between electronic devices originated which revolutionized the global community into a new world of computer networks called internet. The users identified the potential usage of this method presently known as email and started using it as the means of communication and marketing. But the competence of this method was lessened by the wide spread proliferation of spam. Researchers have come up with many proposals and tools to fight against spam. But the dynamic nature of spam makes the tools ineffective and raises the requirement for developing a filter that is to be successful over time in identifying spam. Hence spam filtering is a particularly exigent machine learning task as the data distribution and concept being learned changes over time. This paper explores this phenomenon called concept drift seen in email datasets and proposes a new framework in identifying the strategies for developing spam detection systems.