Spam Email Classifier

The project uses a multi-layer neural network to classify emails as spam or legitimate. The classifier has 98.6% accuracy, and a 97% F1 score.

Key Features

• Text preprocessing with NLTK for cleaning and normalization
• TF-IDF vectorization for text representation
• Multi-layer neural network using PyTorch
• REST API for email classification

Model Architecture

The model uses a multi-layer neural network with:

• Input layer matching feature dimensions
• Three hidden layers with ReLU activation and Batch Normalization
• Dropout regularization to prevent overfitting
• Output layer with sigmoid activation for binary classification

Evaluation Results

The model's performance can be seen through the following visualizations:

Confusion Matrix

The confusion matrix shows:

• True Negatives (top left): Correctly identified non-spam emails
• False Positives (top right): Non-spam emails incorrectly flagged as spam
• False Negatives (bottom left): Spam emails that were missed
• True Positives (bottom right): Correctly identified spam emails

Probability Distribution

This histogram shows how the model distributes probability scores for spam and non-spam emails.

API

To run the project's REST API use the following command:

python app/api.py

Emails that need to be classified can be routed in a POST request to the /predict endpoint with the email provided in the request body.