README.md

metadata

datasets:
  - AbdulHadi806/mail_spam_ham_dataset
language:
  - en
metrics:
  - f1
base_model:
  - distilbert/distilbert-base-uncased

Deep Learning Project: Spam Detection with DistilBERT

This repository contains the code and resources for the Deep Learning project on Spam Detection.

Project Structure

• mail_data.csv: The dataset used for training and evaluation.
• eda_script.py: Script for Exploratory Data Analysis and visualization.
• train_model_hf.py: Main training script using Hugging Face Trainer and DistilBERT.
• evaluate_final.py: Script for final evaluation from the best model checkpoint.
• eda_plots.png: Visualizations generated during EDA.
• results.txt: Detailed evaluation metrics and confusion matrix.
• Deep_Learning_Project_Report.pdf: The final project report (15-17 pages equivalent).

Requirements

• Python 3.11+
• PyTorch
• Transformers
• Datasets
• Scikit-learn
• Pandas
• Matplotlib
• Seaborn
• Accelerate

How to Run

0. Make sure you have all requirements downloaded. In case of errors while running the code, try installing the dependencies in requirements.txt in a fresh environment.
1. EDA: Run python3 eda_script.py to see the data distribution.
2. Training: Run python3 train_model_hf.py to fine-tune the DistilBERT model.
3. Evaluation: Run python3 evaluate_final.py to get the final performance metrics.

Results

The model achieves 99.10% accuracy on the test set with an F1-score of 96.58% for the spam class.