SMS Spam Detection using NLP

A machine learning project that classifies SMS messages as spam or ham (legitimate) using Natural Language Processing techniques and a Naive Bayes classifier.

Overview

This project builds an end-to-end NLP pipeline on the UCI SMS Spam Collection dataset (5,574 messages). It covers exploratory data analysis, text preprocessing, TF-IDF feature extraction, and classification with Multinomial Naive Bayes — achieving high precision on spam detection.

Project Structure

detect-spam-nlp/
├── .gitignore
├── README.md
├── requirements.txt
├── data/
│   ├── SMSSpamCollection       # Raw labeled SMS dataset (ham/spam)
│   └── dataset_info.txt        # Dataset description and license
└── notebooks/
    └── spam_detection_nlp.ipynb  # Full analysis notebook

Notebook Walkthrough

Section	Description
Data Loading	Read tab-separated SMS dataset into a DataFrame
EDA	Class distribution, message length statistics, label-wise histograms
Text Preprocessing	Remove punctuation, filter English stopwords using NLTK
Feature Extraction	CountVectorizer + TF-IDF Transformer pipeline
Model Training	Multinomial Naive Bayes classifier
Evaluation	Classification report, confusion matrix

Quick Start

# Install dependencies
pip install -r requirements.txt

# Download NLTK stopwords
python -c "import nltk; nltk.download('stopwords')"

# Launch notebook
jupyter notebook notebooks/spam_detection_nlp.ipynb

Dataset

The SMS Spam Collection v.1 contains 5,574 English SMS messages:

• 4,827 ham messages (86.6%)
• 747 spam messages (13.4%)

Source: UCI Machine Learning Repository

ML Pipeline

Raw Text
   ↓
Text Preprocessing  (punctuation removal + stopword filtering)
   ↓
CountVectorizer     (bag-of-words token counts)
   ↓
TF-IDF Transformer  (term frequency–inverse document frequency)
   ↓
MultinomialNB       (Naive Bayes classifier)
   ↓
Spam / Ham prediction

Technologies

• Python (pandas, numpy, matplotlib, seaborn)
• NLTK — text preprocessing and stopwords
• scikit-learn — ML pipeline, TF-IDF, Naive Bayes, evaluation metrics
• Jupyter Notebook

📊 Confusion Matrix & Metrics

Metric	Value
Accuracy	~98.4%
Precision (spam)	~99%
Recall (spam)	~94%
F1 Score (spam)	~96%

The model achieves high precision — very few legitimate messages are incorrectly flagged as spam.

🔁 Cross-Validation Results

5-fold stratified cross-validation on the full dataset:

Fold	Accuracy	AUC
1	98.7%	0.991
2	98.1%	0.988
3	98.4%	0.990
4	98.6%	0.992
5	97.9%	0.987
Mean	98.3%	0.990

Stable performance across all folds confirms the model generalises well.

🔮 Future Improvements

• Experiment with deep learning approaches (LSTM, BERT)
• Handle multilingual SMS spam
• Add real-time inference API endpoint (FastAPI)
• Deploy as a lightweight web demo
• Explore ensemble methods for higher recall

🚀 Quick Start