Improving Spam Detection Using Neural Networks Trained by Memetic Algorithm

Abstract

In this paper we train an Artificial Neural Network (ANN) using Memetic Algorithm (MA) and evaluate its performance on the UCI spambase dataset. The Memetic algorithm incorporates the local search capacity of Simulated Annealing (SA) and the global search capability of Genetic Algorithm (GA) to optimize the parameters of the ANN. The performance of the MA is compared with traditional GA in training the ANN. We further explore the different parameters, mechanisms and architectures used to optimize the performance of the network and attain a practical balance between the global genetic algorithm and the local search technique. Classification using ANN trained by MA yielded better results on the spambase dataset compared with other algorithms reported in literature.