📩 SMS Spam Classification using Machine Learning

🚀 Internship Project | NLP & Text Classification

---

📌 Project Overview

This project implements an SMS Spam Detection System using
Natural Language Processing (NLP) and a Multinomial Naive Bayes classifier.

The model classifies SMS messages into:

• ✅ HAM – Legitimate messages
• 🚫 SPAM – Promotional or unwanted messages

The system is trained on the popular SMS Spam Collection Dataset and achieves high accuracy with excellent precision.

---

📂 Dataset Information

• Dataset Name: SMS Spam Collection
• Total Messages: 5572

Class Distribution

• HAM: 4825 messages (86.59%)
• SPAM: 747 messages (13.41%)

Columns:

• label → ham / spam
• message → raw SMS text

---

🧰 Technologies Used

• Python 3.x
• Pandas
• NLTK
• Scikit-learn
• Matplotlib

---

🧹 Text Preprocessing

Each SMS message undergoes the following preprocessing steps:

• Convert text to lowercase
• Remove punctuation
• Remove digits
• Remove English stopwords (NLTK)
• Store cleaned text in a new column

Example

"Free entry in 2 a wkly comp!!!" ↓ "free entry wkly comp"

---

🔤 Feature Extraction (TF-IDF)

Technique Used: TF-IDF Vectorization

Why TF-IDF?

• Converts text into numerical vectors
• Highlights important words
• Reduces impact of very common words

TF-IDF Shapes

• Training Set: (4457, 7431)
• Testing Set: (1115, 7431)

---

🤖 Machine Learning Model

• Model: Multinomial Naive Bayes

Why Naive Bayes?

• Works extremely well for text classification
• Fast and memory efficient
• Probabilistic interpretation

---

✂️ Train–Test Split

• Training Data: 80%
• Testing Data: 20%
• Stratified split to preserve class balance

---

📊 Model Evaluation

Performance Metrics

Metric	Score
Accuracy	0.9641
Precision	1.0000
Recall	~0.75
F1 Score	0.8450

Interpretation

• 🔥 Very high precision → Almost no HAM classified as SPAM
• ✅ Good recall → Most SPAM messages are detected
• ⚖️ Balanced F1 score

---

📨 Sample Predictions

The model displays 5 random test predictions, showing:

• Original message
• Actual label
• Predicted label

Example: Message : hey tmr meet bugis Actual Label : HAM Predicted Label : HAM

---

📈 Data Visualizations

The project generates and saves the following plots:

📌 Class Distribution Plot

• Shows HAM vs SPAM message counts
• Saved as:
  results/class_distribution.png

📌 Top Spam Indicator Words

• Displays most influential words for SPAM classification
• Extracted from Naive Bayes log probabilities
• Saved as:
  results/top_spam_words.png

---

▶️ How to Run the Project

1️⃣ Install dependencies

pip install requirements

2️⃣ Download NLTK stopwords (one-time)

import nltk
nltk.download("stopwords")

#3️⃣ Run the model

python model.py

📌 Key Highlights

• Complete Machine Learning pipeline
• Real-world NLP dataset
• Clean and effective text preprocessing
• Strong evaluation metrics (Accuracy, Precision, Recall, F1-score)
• Clear and meaningful visualizations
• Well-structured, readable, and commented code

🧠 Future Improvements

• Add confusion matrix visualization
• Experiment with Logistic Regression and SVM
• Perform hyperparameter tuning for better performance
• Deploy the model using Flask or Streamlit

👨‍💻 Author

Arnab Datta
Internship Project – Machine Learning & NLP