Comparative Study of the Detection of Malicious URLs Using Shallow and Deep Networks

Abstract

In the modern epoch, all information is easily accessible through websites and due to this reason people rely completely on online resources. On the contrary to its advantages, privacy and security in online media are the main concern worldwide because of the rise in phishing attacks launched online. The number of phishing websites increases every month targeting more than 450 brands, as per the reports published by anti-phishing working groups(APWG). Traditionally blacklists are used to detect the URL attacks. But with the exponential increase in the number of phishing websites, this method has its own limitations and it also fails to detect newly generated phishing URLs which can be solved using machine learning or deep learning techniques. Here we present a comparative study between classical machine learning technique - logistic regression using bigram, deep learning techniques like convolution neural network(CNN) and CNN long short-term memory(CNN-LSTM) as architectures used to detect malicious URLs. On comparison CNN-LSTM gave the best accuracy of about 98% for the classification of phishing URLs.