Web Hazard Identification and Detection Using Deep Learning - A Comparative Study

Abstract

Surfing the internet has become an integral part of our day-to-day life. This has become the potential source of intruder attacks. Hazard is
cybercriminal posed threat, the simple example for the same is creating malicious URL to pose phishing attack and to gain access to user’s
personal information. The consequences include identity theft, other types of frauds like malware injection onto the computing devices.
Malicious URL is a link that redirects the user to a fraudulent web page. Recognition of such malicious URLs is a prolonging problem since
machine learning (ML) has evolved. There have been ML classifier like random forest (RF) and deep learning (DL) classifiers such as
Convolutional Neural Network (CNN), Back Propagation Neural Networks (BPNN) and Long Short-Term Memory (LSTM) which may
address this classification problem of segregating URLs into malicious and normal. Still these techniques are not sufficient to protect the
internet users and requires a robust model that will distinguish between the normal and malicious web pages. This paper introduces a
comparative study about ML and DL techniques in classification of URLs as malicious and normal in the given dataset. Among the implemented
techniques BPNN gave an optimal accuracy of 96.86%