CyberGuard-AI

By Aditya Singh & Mohak Gupta

Introduction

This project involves building a cybercrime classification model using a BERT-based architecture. The model is designed to classify text data into various categories and subcategories related to cybercrime.

Dependencies

The following libraries are required to run the project:

• pandas: For data manipulation and analysis.
• numpy: For numerical operations.
• torch: The PyTorch library for building and training the model.
• transformers: Hugging Face's library for state-of-the-art natural language processing models.
• nltk: For natural language processing tasks.
• sklearn: For machine learning utilities and metrics.

Dataset Preparation

The CybercrimeDataset class is used to prepare the data for training. It handles tokenization and encoding of the input text.

Model Architecture

The CybercrimeClassifier class defines the BERT-based model. It consists of:

• BERT Backbone: The model uses a pretrained BERT model to extract features from the input text.
• Dropout Layer: Applied to reduce overfitting.
• Hidden Layers: Fully connected layers to process BERT embeddings.
• Output Layers: Separate classifiers for predicting categories and subcategories of cybercrime.

Training Process

The model is trained using the train method in the CybercrimeNLP class. The training involves:

• Preparing the data into batches.
• Running a loop for a specified number of epochs to update model weights.
• Calculating the loss and updating the model using backpropagation.

Evaluation

The evaluate method assesses the model's performance on a validation dataset. It calculates the average loss and can be expanded to include metrics such as accuracy and F1 score.

Inference

The predict method allows the model to make predictions on new text. It outputs the predicted categories along with confidence scores.

Saving and Loading the Model

The model can be saved to disk using the save_model method and loaded back using the load_model method, ensuring that the model state and label encoders are preserved.

Example Usage

In the main function, the model is trained and tested on sample data. Here’s an example of how to make a prediction:

test_text = "I received a call from an unknown number..."
prediction = classifier.predict(test_text)
print(prediction)


### Conclusion
```markdown
## Conclusion
This project demonstrates the implementation of a BERT-based model for cybercrime text classification, highlighting the potential of deep learning in tackling cybersecurity issues.