A Study for ANN Model for Spam Classification

Abstract

The classical way of detecting spam emails based on the signature is not very effective in recent days due to the huge uses of emails in various activities. The online recommendation and push emails make the spam detection job very complex and tedious. Machine learning happens to be a widely used approach for automated email spam detection. Among various machine learning algorithms, Artificial Neural Network (ANN) is gaining popularity due to its powerful approximation and generalization characteristic. The effectiveness of the email spam classifier is heavily dependent on the learning capability of ANN. In our work, we have developed a BP and a BP+M model to do the spam classification and find the accuracy of classification. We have compared the two models so that we can conclude that the BP+M model gives the same or better result than the BP model using fewer epochs. Though state-of-the-art and classical learning algorithms like backpropagation (BP) and backpropagation with momentum (BP+M) are very popular and well researched, it is understood that often it gets trapped in local optima. In our future work, we can use recent optimization techniques like SGO which can elevate the results and can eradicate the drawbacks of BP and BP+M model. After thorough simulations and results analysis, we conclude that backpropagation + momentum optimized ANN provides superior classification results than BP optimized ANN.