Spam Email Detection with Affect Intensities using Recurrent Neural Network Algorithm

Abstract

A large number of email users triggers an increase in the occurrence of spam in emails to gain benefits for some parties but harm others and also email users. Spam emails usually contain advertisements or criminal acts such as phishing which implicitly contain human emotions in them. It is quite difficult and takes time to differentiate between a large number of spam and ham emails. This problem can be overcome by using deep learning technology. One of which is a neural network that can classify spam emails. This paper uses the spam and ham Enron email corpus dataset. This study will add emotional features in extracting its features. The steps taken include text preprocessing, feature extraction using tf-idf, and lexicon-based emotion features, followed by classification using RNN to detect spam in emails. A comparison with other methods is also provided by comparing the proposed method to Naïve Bayes and Support-Vector Machine (SVM) algorithm based on precision and accuracy. In addition, this study also compares the effect of using affect intensities on the performance of algorithms. The results show that RNN outperforms other methods by showing the highest accuracy 99% and the precision of 99.1%. Adding effect intensities to the model would increase the model recognition results.