Phishing Website Detection Using Fast.ai library
Jayesh V Jawade, Soma Ghosh
Abstract
In recent years, we see an enormous amount of increase in online activities whether it be for using an internet banking services or for using social media or for education purpose. On other side, most of the peoples did not focus on the privacy because either they are not well aware about the harm of privacy breach or they simply neglect the security part since security comes at the cost of some constraints. Because of these reasons we are witnessing a rapid increase in cyber security attacks. Now-a-days one of the top ten cyber-attacks is phishing attack. Phishing attack is an attempt to mimic the legitimate websites to obtain or gain the confidential data such as credit or debit card details, username and password, etc. The problem of phishing attack detection is addressed so many times. Recent advancements in this domain is classification using the deep learning techniques. However the results of deep learning technique highly depends on the architecture of the system and parameter tuning of learnable parameters. Most of the previous techniques requires high computational power systems to train the deep learning or neural network models. Hence, such approaches requires more time to train the model. In this paper, we propose a convolutional neural network architecture using fast.ai library across GPU which solves the problem of traditional neural networks. By applying the proposed method, we were able to improve the phishing detection accuracy and training time when compared to other existing approaches. We use the ISCX-URL-2016 dataset to train our model for binary classification as phishing and legitimate. We got the accuracy of 99% for phishing URL's detection.