SMS Spam Classifier

📩 SMS Spam Classifier using Naive Bayes

This project builds a text classification model to detect whether an SMS message is spam or ham (not spam) using Naive Bayes and TF-IDF vectorization. It is a classic NLP use case and demonstrates how probabilistic classifiers can be used to flag unwanted messages.

---

📌 Objective

To classify SMS messages as spam or ham using machine learning, and evaluate model performance through classification metrics and confusion matrix analysis.

---

📂 Dataset

• Source: Kaggle - SMS Spam Collection
• Total Samples: 5,572 messages
• Target Variable:
  • spam: Unwanted messages
  • ham: Legitimate messages
• Format:
  • label: spam/ham
  • message: the text content of the SMS

---

🔍 Exploratory Data Analysis (EDA)

• Checked class distribution (spam vs ham)
• Visualized frequent spam/ham words using word clouds
• Observed that spam messages tend to use promotional terms, numbers, and URLs

---

⚙️ Preprocessing

• Removed punctuation, stopwords, and converted text to lowercase
• Encoded labels: ham → 0, spam → 1
• Applied TF-IDF Vectorizer to transform messages into numerical features
• Split data into training and test sets with stratification

---

🧠 Model Used

• Multinomial Naive Bayes: Ideal for discrete features like word counts or TF-IDF scores
• Trained using scikit-learn’s MultinomialNB

---

📈 Model Evaluation

Metric	Ham (0)	Spam (1)	Interpretation
Precision	0.95	1.00	No ham misclassified as spam
Recall	1.00	0.62	38% of spam messages were missed
F1-Score	0.97	0.76	Spam recall could be improved
Accuracy	–	–	95.16% overall

Confusion Matrix:

[[903   0]
 [ 50  81]]

• ✅ True Ham: 903
• ❌ Missed Spam: 50
• ✅ Correct Spam: 81

---

🔍 Sample Prediction

predict_message("Congratulations! You've won a free vacation! Reply NOW to claim.")
# Output: 'Spam'

The model takes any SMS input and returns a prediction: Spam or Ham.

---

💻 How to Run

# 1. Clone or download the repo
# 2. Install dependencies
pip install pandas numpy scikit-learn matplotlib seaborn wordcloud

# 3. Run the Jupyter Notebook
# 4. Use predict_message() function to test your own texts

---

🚀 Future Enhancements

• Improve recall for spam class by tuning alpha or using other classifiers
• Experiment with n-grams or additional features
• Deploy with Streamlit or Flask for interactive UI
• Use deep learning (LSTM, BERT) for more advanced performance

---

👤 Author

• GitHub: archangel2006