Improving the Performance of Heuristic Spam Detection using a Multi-Objective Genetic Algorithm
James Dudley · 2007
With today’s reliance on email, spam is not just annoyance, but presents a real cost to companies and individuals. Several methods of spam detection exist, but each has certain weaknesses. Heuristic spam filters address these weaknesses by using several different methods at once; however the complexity of these filters makes it difficult to optimise their performance. Complicating the problem is that spam detection has more than one objective: as well as catching as much spam as possible, a spam detector must limit the amount of legitimate email that is incorrectly flagged as spam. This dissertation investigates the use of genetic algorithms to optimise heuristic spam filter performance. The use of multi-objective genetic algorithms to find a set of trade-off solutions is also examined. This is done by implementing both a single objective and multi-objective genetic algorithm over SpamAssassin, which is a popular heuristic spam filter. Positive results are obtained, with both algorithms outperforming the current method of optimisation used by SpamAssassin, and the multi-objective algorithm producing a front of solutions that allows a solution to be selected based on the desired trade-off between the two objectives of catching spam and losing legitimate email.