Malicious URL Classification Using Deep Neural Network

Abstract

One of the most serious cybersecurity threats has been discovered as malicious URL classification. People are duped into believing they are visiting a respectable website, and they are persuaded to provide their credentials by malicious websites. Many cyber crimes, such as phishing, cyberbullying, spamming, and malware, are founded on the hosting of dangerous URLs. The existing methods are slow and rely on manual feature engineering. In this research, we look at Convolutional Neural Networks, Long Short Term Memory, and a hybrid model of CNN followed by LSTM, and we propose a new strategy based on the results. Character embedding, in which each character is treated as a token, and word embedding, in which each word is treated as a token, are both used in our technique. In comparison to the word embedded hybrid mode1(0.988), the character embedded hybrid model has a higher accuracy (.996).