Phishing URL Detection: ML & Transformer-Based Text Embeddings

Overview

This project explores the problem of detecting phishing websites using machine learning techniques. The goal is to build models capable of distinguishing between benign and phishing URLs by leveraging a feature-engineered dataset containing structural and lexical characteristics extracted from URLs.

The work is based on a replication of the research study “A Feature-Engineered Dataset of Benign and Phishing URLs for Machine Learning and Large Language Models Evaluation.” The objective is to reproduce the experimental setup described in the paper and evaluate whether similar model performance can be achieved using the provided dataset.

---

More Details

The project currently includes two primary notebooks:

1. Replication Study

The notebook phishing_url_detection.ipynb reproduces the experimental methodology described in the original research paper. It trains and evaluates two approaches:

• Random Forest classifier trained on engineered URL features
• MiniLM embeddings + Logistic Regression trained on raw URL text

This notebook contains detailed explanations of the dataset, the experimental setup, model training, and comparisons with the results reported in the original paper.

2. External Model Evaluation

The notebook phishing_url_model_external_evaluation.ipynb evaluates the generalization ability of the trained MiniLM + Logistic Regression model on a separate dataset containing only raw URL strings. Goal: To help assess how well the transformer-based approach performs on unseen URL distributions.

---

Additional resources:

• Replication Study Notebook Kaggle
• Original Dataset Kaggle
• Original Dataset Mendeley
• Original Paper A Feature-Engineered Dataset of Benign and Phishing URLs...

How to Run

1. Clone the repository.
2. Ensure the dataset file is available in the working directory (download link above).
   • You may need to adjust the file path in the notebook accordingly.
3. Open terminal/cmd and navigate to the project directory.
4. Install the required dependencies using pip or Conda.
5. Run the Jupyter notebook to execute the code and reproduce the results.

git clone https://github.com/SahandNamvar/phishing-url-detection.git
cd phishing-url-detection
pip install -r requirements.txt # Or Conda environment setup
jupyter notebook phishing_url_detection.ipynb