Spam Detection with Stacking Classifier

A sophisticated machine learning web application that detects spam messages using an ensemble stacking classifier. Built with FastAPI backend and a modern, responsive frontend.

Features

• Advanced ML Model: Stacking classifier combining MultinomialNB, LinearSVC, SGDClassifier, and RandomForestClassifier
• Optimized Performance: RandomizedSearchCV hyperparameter tuning for best results
• RESTful API: FastAPI backend for fast, reliable predictions
• Modern UI: Professional, responsive web interface.
• Real-time Analysis: Instant spam detection with visual feedback
• High Accuracy: Optimized for recall to minimize false negatives

Model Performance

• Accuracy: ~98%
• Precision: ~88%
• Recall: ~93%
• F1 Score: ~90%

Note: Metrics based on test set evaluation

Screenshots

Predicting SPAM

Predicting HAM

Project Structure

spam-detection-with-stacking-classifier-machine-learning/
│
├── data/
│   ├── spam_dataset.csv          # Original dataset
│
│── spam_classifier.pkl           # Trained model
├── notebooks/
│   ├── spam_detection.ipynb
│
│── main.py                       # FastAPI application
│── index.html                    # Frontend HTML/CSS/JS
│
│
├── .gitignore                    # Git ignore rules
├── README.md                     # This file
└── LICENSE                       # MIT License 

Installation

1. Clone the repository

   git clone this repo
   cd spam-detection-with-stacking-classifier-machine-learning

2. Create a virtual environment

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Run the application

   uvicorn main:app --reload

4. Access the web interface Open your browser and navigate to http://localhost:8000

Requirements

fastapi==0.104.1
uvicorn==0.24.0
scikit-learn==1.3.2
pandas==2.1.3
numpy==1.26.2
pydantic==2.5.2
nltk

Model Architecture

Base Models

• MultinomialNB: Naive Bayes with Laplace smoothing
• LinearSVC: Support Vector Classification with balanced class weights
• SGDClassifier: Stochastic Gradient Descent with modified Huber loss
• RandomForestClassifier: Ensemble of decision trees with 200 estimators

Meta-Learner

• LogisticRegression: Combines predictions from base models

Feature Engineering

• TfidfVectorizer: N-gram based text vectorization
  • Max features: 8000
  • N-gram range: (1,1) to (1,3)
  • Sublinear TF scaling

Hyperparameter Tuning

RandomizedSearchCV was used with:

• n_iter: 100 iterations
• cv: 5-fold cross-validation
• scoring: Recall (to minimize false negatives)
• n_jobs: Parallel processing enabled

Tuned Parameters

Component	Parameter	Range
TF-IDF	max_features	3000-11000
TF-IDF	ngram_range	(1,1) to (1,3)
MultinomialNB	alpha	0.01-1.0
LinearSVC	C	0.1-2
RandomForest	n_estimators	50-300
LogisticRegression	C	0.01-5

🌐 API Endpoints

POST /predict

Predicts whether a message is spam or not.

Request:

{
  "text": "Your message here"
}

Response:

{
  "prediction": "spam" | "not spam"
}

GET /

Returns the web interface (HTML page).

📝 License

This project is licensed under the MIT License

👤 Author

Chinmoy Guha

• GitHub: @LT-Ripjaws
• Email: chinmoyguha676z@gmail.com