Phishing URLs
A balanced set of legitimate and fraudulent URLs, designed for phishing detection using automatic analysis methods.
Description
The dataset Phishing URLs contains 96,018 entries split evenly between legitimate URLs and phishing URLs. Each line is represented by a domain, a label (0 for legitimate, 1 for phishing) and a set of characteristics calculated according to metrics from the research article “PhishStorm”.
What is this dataset for?
- Train classification models to detect fraudulent URLs
- Testing web security algorithms in real time
- Evaluate machine learning approaches on balanced binary data
Can it be enriched or improved?
Yes, it is possible to complete this data set with additional characteristics (e.g. WHOIS, IP geolocation, DNS history). Feature engineering techniques or temporal analysis models can also enrich the exploitation of the dataset.
🔎 In summary
🧠 Recommended for
- Cybersecurity projects
- Phishing detection training
- Binary classification
🔧 Compatible tools
- Scikit-learn
- Pandas
- LightGBM
- XGBoost
- TensorFlow
💡 Tip
Use balanced distribution to experiment with different evaluation metrics like AUC, F1-score, and precision/recall.
Frequently Asked Questions
Can this dataset be used to train a model in production?
Yes, as long as you enrich it with more recent data and adapt the features to the context of your web service.
Is the dataset balanced between phishing and non-phishing?
Yes, it contains exactly 48,009 legitimate URLs and 48,009 phishing URLs.
Is it suitable for supervised learning?
In fact, the dataset is labeled and formatted for training classic or advanced supervised models.
