BERT FINETUNED ON PHISHING DETECTION

This model is a fine-tuned version of bert-large-uncased on an phishing dataset, capable of detecting phishing in its four most common forms: URLs, Emails, SMS messages and even websites.

It achieves the following results on the evaluation set:

• Loss: 0.1953
• Accuracy: 0.9717
• Precision: 0.9658
• Recall: 0.9670
• False Positive Rate: 0.0249

Model description

BERT is a transformers model pretrained on a large corpus of English data in a self-supervised fashion. This means it was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of publicly available data) with an automatic process to generate inputs and labels from those texts.

This model has the following configuration:

• 24-layer
• 1024 hidden dimension
• 16 attention heads
• 336M parameters

Motivation and Purpose

Phishing is one of the most frequent and most expensive cyber-attacks according to several security reports. This model aims to efficiently and accurately prevent phishing attacks against individuals and organizations. To achieve it, BERT was trained on a diverse and robust dataset containing: URLs, SMS Messages, Emails and Websites, which allows the model to extend its detection capability beyond the usual and to be used in various contexts.

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 4

Training results

Framework versions

• Transformers 4.34.1
• Pytorch 2.1.1+cu121
• Datasets 2.14.6
• Tokenizers 0.14.1