Convolutional Neural Networks and Long Short Term Memory for Phishing Email Classification

Abstract

The focus on this work is on classifying phishing
emails using deep neural networks. Since phishing emails have no
specific characteristic, they are difficult to detect and classify,
and little research has been done on the detection of phishing
emails. In this work, two deep neural networks, Long Short Term
Memory (LSTM), a form of Recurrent Neural Networks (RNN),
and Convolutional Neural Networks (CNN), were compared and
used for classification of phishing emails. RNN is the most used
neural network for text classification. CNNs have also shown to
be effective in text classification. In addition to tuning
hyperparameters, different activation functions and optimizers
are used for comparing the performance of CNN and LSTM on
the basis of accuracy and the ROC-score. LSTM achieved a
higher accuracy than CNN, and overall the Adam Optimizer
performed better than the SGD optimizer. The best parameters
for higher accuracy and ROC-score are also presented.
Keywords: Phishing Email Classification; Convolutional
Neural Networks; Long Short Term Memory; Hyperparameters;
Recurrent Neural Networks; Deep Learning