Classifications of E-MAIL SPAM Using Deep Learning Approaches

Abstract

Spamming is an art used to deceive people and, in most cases, send unwanted messages that can be used by cybercriminals to trick victims and get confidential credentials from the victim. Spamming occurs via email, SMS, social networking websites calling the victim’s phone number, etc. Spamming can exist for various reasons, but it can be used for malicious purposes mainly like trying to forge the victim for gathering the personal information, bank details card details, passwords, and other confidential data about the intended user. To overcome the security breach, the spam messages are classified for the understanding of spam has been done using different methods like Machine Learning and Natural Language Processing. We propose a method to identify the email spam messages and also to classify the message using deep learning approaches such as Recurrent Neural Network (RNN), Long-Short Term Memory (LSTM), and Bilateral Long Short-Term Memory (BLSTM). We have used an email spam dataset with over 5000 spam text samples are used to train our deep learning models. The proposed method separated the sentences from the email into words followed by their root words, and each word is assigned an index number before training. We have used regularization and dropout layers in our models to reduce the chances of overfitting. The proposed method is evaluated and compared with other existing models based on the history of the loss curve, Precision-Recall, and accuracy. Based on the results, it is observed that our Bi-LSTM model produced higher accuracy than other existing systems.