Ways to Evade Spai Filters and Machine Learning as a Potential Solution
Vivek Kumar Chandra, Nikita Shrivastava · 2006
The growth of unsolicited email or junk mail has become a major threat to information security today. Most of the commercial Websites providing email facility today provide some form of programmable automatic filtration of junk mails, typically in the form of a set of rules to dispose of mails based on keywords detected in headers or message body. There are some other methods for the purpose as well, but an adaptive email filtering system based on Bayesian technique that can learn from its user's mail preferences appears to be a better solution. It not only considers the keywords that identify spam, but also words that denote innocent or legitimate mail which makes the filter more efficient . However direct marketers discover new ways to evade even the Bayesian filters, so as to reach the potential customers. The way most commonly adopted by the spammers to mislead the Bayesian filters is injection of out of context text in the email. This paper aims at presenting a critical analysis of the various ways adopted by spammers to dodge the spam filters. Further we explore the Bayesian noise reduction (BNR) technique which attempts to solve this problem by identifying and eliminating the 'out of context' data (so injected by spammers or otherwise) to provide a cleaner classification