Spam SMS Detection using Machine Learning

This project classifies SMS messages as Spam or Not Spam using Naive Bayes algorithms.

Dataset

• SMS Spam Collection dataset
• Stored in data/spam.csv
• Encoded in UTF-8

📊 Dataset Update & Model Improvement

Originally, this project used a standard Kaggle SMS spam dataset. While it performed well on promotional spam, recall was poor on modern scam patterns such as:

• account security alerts
• fake delivery messages
• job and investment scams
• invoice and refund phishing

To address this, a custom dataset was generated containing modern spam and scam patterns.

Models Used

• Gaussian Naive Bayes
• Multinomial Naive Bayes (selected for deployment)
• Bernoulli Naive Bayes

Tech Stack

• Python
• Pandas
• NumPy
• Scikit-learn
• Joblib

Project Structure

spam-detection/
├── spam_sms_detection.ipynb
├── data/
│   └── spam.csv
├── models/
│   ├── mnb.pkl
│   └── vectorizer.pkl
└── requirements.txt

How to Run

1. Install dependencies
   pip install -r requirements.txt
2. Open the notebook
   spam_sms_detection.ipynb
3. Run all cells

NLTK Setup

This project uses NLTK for tokenization. If you face errors related to tokenizers, run:

import nltk
nltk.download('punkt')


### Results
- Recall improved from ~74% → ~99.6%
- Precision remains ~99%
- Model now generalizes better to real-world scam messages

### Note on Limitations
This is a text-only model. Certain messages such as neutral security
alerts or charity requests may still be ambiguous without metadata
(sender, headers, links).

## Future Improvements
- Web app deployment
- API support