Improving Spam Email Filtering Systems Using Data Mining Techniques

Introduction

E-mail is a means of communication which is widely used due to its global accessibility, low-cost and speed of information exchange. This makes it a popular choice of communication in both personal and professional aspects. Due to its popularity, spammers incline towards using email to send spam. Email spam has become one of the major problems of today’s Internet, causing financial losses and data breaches to individual users as well as the organizations. Email spam is any unrequested email sent to a group of recipients. It can also include executable attachments/malware. Email spam is defined as any email that satisfies the following criteria (Rathi and Pareek, 2013):

• 1.

  Anonymity: The sender’s name and email address are unknown to the receiver.

• 2.

  Mass Mailing: The sender sends the email to a large group of recipients.

• 3.

  Unsolicited: The email is unrequested by the recipients.

Spam emails are a nuisance and adversely affect the productivity by consuming users’ time to filter out the irrelevant emails, filling up the mailboxes, bury important emails, wasting the bandwidth etc. Moreover, according to the IBM Threat Intelligence Index 2017 (IBM, 2017), spam emails are one of the primary channels used by cybercriminals to spread malware. Users receive hundreds of spam emails every single day. These emails are sent from new email addresses with different content which makes it almost impossible to filter these spam emails using the traditional methods of blacklist and whitelist. A spam filter is one of the most important techniques to detect and prevent unwanted emails from getting into the users’ inbox. It looks for certain criteria to judge the relevance of the email. The simplest version of a spam filter can be configured to identify certain groups of words occurring in the subject of the email and excluding them from the user’s inbox. This is an overly simplistic and ineffective technique of spam filtration as the probabilities of false positives and false negatives are very high.

In 2004, Bill Gates famously said, “Two years from now, spam will be solved” and today 15 years later, spam is still a huge and ever-increasing problem. Therefore, it is clear that further research is required in order to study and improve spam filtering algorithms. This research is to study data mining techniques to solve the problem under study.

Over the years, email spam has become a significant problem faced by users on the Internet. These emails waste time, space and bandwidth. Spam emails are a threat to the privacy of the users. It has become an important security issue as it is also a means for propagating threats like viruses, worms, malware etc. According to recent statistics, 85% of the received emails are spams (CISCO, 2019). Despite all the research in the area of spam email filtering, on average the amount of email spam is increasing by 4% every month. Numerous methods have been developed to solve this problem, but spammers have found ways to evade the spam filters and get their emails into users’ inboxes. To address this issue, this research will focus on increasing the accuracy of spam filters while reducing the number of false positives.

The following research questions have been formulated:

• 1.

  What are the best possible techniques currently to filter spam emails and reduce false positives using data mining?

• 2.

  How can the accuracy of the current data mining algorithms be improved for spam filtering?

Accordingly, the following are the chapter objectives:

• 1.

  To analyze the effectiveness of current data mining techniques for spam filtering.

• 2.

  To propose an algorithm for improving the accuracy of spam filters in order to reduce the rate of false positives.

• 3.

  To validate the proposed algorithm using multiple email datasets.

• 4.

  To compare the proposed algorithm with other similar existing techniques for spam filtering.