Voting-based Classification for E-mail Spam Detection

The problem of spam e-mail has gained a tremendous amount of attention. Although entities tend to use e-mail spam filter applications to filter out received spam e-mails, marketing companies still tend to send unsolicited e-mails in bulk and users still receive a reasonable amount of spam e-mail despite those filtering applications. This work proposes a new method for classifying e-mails into spam and non-spam. First, several e-mail content features are extracted and then those features are used for classifying each e-mail individually. The classification results of three different classifiers (i.e. Decision Trees, Random Forests and k-Nearest Neighbor) are combined in various voting schemes (i.e. majority vote, average probability, product of probabilities, minimum probability and maximum probability) for making the final decision. To validate our method, two different spam e-mail collections were used.