A Long Short-Term Memory-based Phishing Detection on URL

ABSTRACT

The rapid evolution of cyberspace and the advent of new technologies have greatly impacted our daily lives. However, these advancements also present vulnerabilities that malicious actors exploit. Phishing, a common cyber threat, involves creating deceptive traps that can lead to severe consequences like financial loss and blackmail. Traditional detection methods, such as blacklisting, whitelisting, the Cantina approach, heuristics, and visual similarity, are limited as attackers quickly adapt to bypass them. This research addresses this challenge using Long Short-Term Memory (LSTM), a Recurrent Neural Network (RNN) type in deep learning. LSTM, which mimics the brain's sequential processing and includes memory cells and gates, improves the network's ability to learn and recognize patterns over time. Using a dataset from the University of Brunswick and PhishTank, the study preprocessed URL characters into a one-hot encoded sequence for LSTM input. Results show that the LSTM model detects phishing attacks with 98.2% accuracy, offering a scalable, fast, and efficient solution.

Keywords: Phishing, URL, cyberspace, Recurrent Neural Networks, Long Short Term Memory