Modeling Spammer Behavior: Artificial Neural Network vs. Naïve Bayesian Classifier

The exponential growth of spam emails in recent years is a fact of life. Internet subscribers world-wide are unwittingly paying an estimated €10 billion a year in connection costs just to receive "junk" emails, according to a study undertaken for the European Commission. Though there is no universal definition of spam, unwanted and unsolicited commercial email as a mass mailing to a large number of recipients is basically known as the junk email or spam to the internet community. Spams are considered to be potential threat to Internet Security. Spam's direct effects include the consumption of computer and network resources and the cost in human time and attention of dismissing unwanted messages. More importantly, these ever increasing spams are taking various forms and finding home not only in mailboxes but also in newsgroups, discussion forums etc without the consent of the recipients. Overflowing mailboxes are overwhelming users, causing newsgroups and discussion forums to be flooded with irrelevant or inappropriate messages. As a consequence, users are getting discouraged not to use them anymore though these systems can provide numerous benefits to them. Combating spam is a difficult job contrast to the spamming. Millions of spammers around the world are engaged in spreading spams with ever changing tricks and tactis to circumvent the filters deployed by the mailbox providers. As spammers are paid per volume for thier job, they invest thier best effort in reaching everyone by all possible ways. No antispamming technique is hundred percent accurate for spam problem. Antispamming techniques try to make a trade-off between rejecting legitimate e-mail vs. not rejecting all spam, and the associated costs in time and effort. The simplest and most common approaches are to use filters that screen messages based upon the presence of common words or phrases common to junk e-mail. Other simplistic approaches include blacklisting and whitelisting. • Blacklisting technique automatically rejects messages received from the addresses of known spammers. • Whitelisting accepts messages received from known and trusted correspondents only. The major flaw in the first two approaches is that it relies upon complacence by the spammers by assuming that they are not likely to change (or forge) their identities or to alter the style and vocabulary of their sales pitches. Whitelisting risks the possibility that the