Tools
Tools
Ikelia, Emmanuel Ugochukwu (2025) FDFN-SA: The Lightweight Phishing detection system for endpoint devices. Masters thesis, Dublin, National College of Ireland.
Preview |
PDF (Master of Science)
Download (970kB) | Preview |
Preview |
PDF (Configuration Manual)
Download (329kB) | Preview |
Abstract
Phishing is a type of cyber-attack that involves luring people into clicking on deceptive URLs designed to evade traditional phishing detection tools, resulting in a critical need for lightweight and more innovative anti-phishing solutions that can be deployed on endpoint devices. Over the past decade, security experts have increasingly turned to employing Artificial Intelligence-based solutions, which have proven to be more effective, but often come with the requirement of high computing costs that are not suitable for deployment on endpoint devices. To overcome these limitations, we introduce the FDFN-SA, short for Fuzzy-Driven Fusion Network with a Sparse Autoencoder, which is a lightweight, endpoint deployable version of the FDN-SA phishing detection model. This lightweight design was achieved by trimming the heavy inference steps and introducing an additional character-level branch to tackle obfuscation techniques, thereby improving detection capability. The resulting model utilizes only 5,676 parameters capable of delivering a decision within 100 milliseconds. Despite being lightweight, when trained and tested using the same dataset as the original FDN SA and another large language model (LLM), our proposed model achieved an accuracy of 94%, outperforming the FDN-SA’s 92% accuracy, which required 156 milliseconds during inference. Additionally, our proposed model achieved 96% accuracy, compared to the 98% achieved by the LLM. The result is a model that can handle modern phishing tactics while remaining lightweight enough for deployment on edge devices. To encourage reproducibility, the code and dataset used in developing our model have been made publicly available.
Actions (login required)
 |
View Item |