Comparação de desempenho entre máquina de vetor de suporte e comitê de redes neurais artificiais para classificação de spam

Abstract

The e-mail is one of the most popular communication tool. However, it is not uncommon to find undesired messages in our electronic mail boxes. These messages are known as spams. In a context where more and more messages are sent and received over the world, computational techniques to filter spams have increasingly importance. This paper aims to present two classifiers to filter e-mail messages, which is to identify whether a particular email is spam or not. It was used two machine learning techniques: Support Vector Machine (SVM) and Artificial Neural Network (ANN). For the ANN classifier, we used five different ANNs with Resilient Propagation (Rprop) learning algorithm (Backpropagation variation) and distinct architectures (layers) and settings (learning rate and number of iterations), forming a committee of networks. The two classifiers showed average rates of accuracy of 91.3% and 93.6% respectively. The SVM showed superior results compared to a single specific ANN (depending on architecture), but it was lower in the performance of a committee of ANNs. Furthermore, SVM is faster in training than the committee of ANN.