adhit-420/bert-phishing-classifier

Model Card: Fine-Tuned Phishing URL Classifier

Model Details

Model Name: Fine-Tuned Phishing URL Classifier
Base Model: BERT base uncased (Hugging Face Transformers)
Library Used: Transformers
License: MIT
Datasets: Rishik001/URLs

Model Description

This model has been fine-tuned for phishing URL classification. It is trained to classify URLs as either phishing or benign using a dataset of labeled URLs. The fine-tuning process leverages a pre-trained BERT model, adapting it to the domain-specific task of URL classification.

Intended Use

This model is intended for cybersecurity applications, specifically for detecting phishing URLs. It can be used in security systems to filter out malicious links and protect users from cyber threats.

Training Configuration

Hyperparameters:

• Learning Rate: 2e-5
• Per Device Train Batch Size: 16
• Per Device Eval Batch Size: 16
• Number of Training Epochs: 3
• Weight Decay: 0.01
• Evaluation Strategy: Epoch
• Save Strategy: Epoch
• Load Best Model at End: True

Training Performance

Epoch	Training Loss	Validation Loss	Accuracy	AUC
1	0.365300	0.315958	0.868000	0.941
2	0.344800	0.288597	0.882000	0.949
3	0.342000	0.284374	0.880000	0.951

Evaluation Results

The model was evaluated on a test set, and the results are as follows:

• Evaluation Loss: 0.290662
• Evaluation Accuracy: 87.5%
• Evaluation AUC: 0.949
• Evaluation Runtime: 60.608 seconds
• Evaluation Samples per Second: 164.995
• Evaluation Steps per Second: 10.312

Citation

If you use this model in your work, please cite it appropriately:

@misc{phishingurlclassifier,
  title={Fine-Tuned Phishing URL Classifier},
  authors={S Adhit and M Rishikesh},
  year={2025},
  publisher={Hugging Face Transformers}
}