Improved Neural Network Model for Real-Time Phishing Email Detection

Phishing emails represent one of the largest hazards in today's world resulting in billions of unwanted monetary losses. While phishing emails detection methods are constantly being assessed, the current results for those methods are not very good. Additionally, phishing emails records show that phishing emails are escalating at an unmanageable speed each year. Accordingly, we need better phishing detection systems to help with the phishing email threat. In this study, we first researched the modality of an email. Then based on an improved Recurrent Convolutional Neural Networks (RCNN), with multilevel vectors and attention mechanism, we proposed a new phishing email detection model called, by looking at the emails content in parallel, including from the email header, the email body, the character level, and the word level. We also used the unbalanced dataset realistic phishing to legitimate email problem as a basis on if the classifier was effective. The results show we proved the effectiveness of, as it provides a way to filter phishing emails with high degree of confidence to pick out the phishing emails, and filter legitimate emails as little as possible. Overall, that is a promising result to perform better than existing methods and provide assurance of effectiveness of in detecting phishing emails.