attmacah/email-spam-detector

---

license: mit tags: - spam-detection - scikit-learn - multinomial-nb - nlp - english

🛡️ SMS & Email Spam Detector

This repository contains a machine learning model designed to classify text messages (SMS or emails) as Spam or Ham (Normal).

📊 Model Details

• Algorithm: Multinomial Naive Bayes (MultinomialNB)
• Feature Extraction: TF-IDF Vectorizer (TfidfVectorizer)
• Dataset: UCI SMS Spam Collection (5,572 real-world messages)
• Accuracy: ~98% on test data

🧠 Deep Dive: Algorithms & Pipeline Architecture

The application relies on a robust Natural Language Processing (NLP) pipeline combined with probabilistic machine learning.

1. Feature Extraction: TF-IDF (TfidfVectorizer)

Raw text cannot be processed directly by mathematical models. The TF-IDF vectorizer converts text messages into weighted numerical matrices:

• Term Frequency (TF): Measures how frequently a word appears in a message.
• Inverse Document Frequency (IDF): Downweights common words (like "the", "is", "to") that appear across all messages and boosts rare, highly discriminative words (like "win", "free", "prize", "click").

2. Classification Algorithm: Multinomial Naive Bayes (MultinomialNB)

• What it is: A probabilistic learning algorithm based on Bayes' Theorem, specifically tuned for text data where features represent word counts or frequencies (multinomial distribution).
• How it works: It calculates the probability of a message belonging to a class (Spam vs. Ham) given the words present in it, assuming independence between word features.
• Why we chose it: It is exceptionally fast, lightweight, highly accurate for text classification tasks, and performs remarkably well with small-to-medium datasets without requiring heavy resources.

---

🚀 Files Included

• spam_model_v2.pkl: The trained Scikit-Learn classification model.
• tfidf_vectorizer_v2.pkl: The fitted TF-IDF vectorizer for text preprocessing.

💻 How to Use

You can load and use the model in your Python project like this:

import joblib

# Load model and vectorizer
model = joblib.load('spam_model_v2.pkl')
vectorizer = joblib.load('tfidf_vectorizer_v2.pkl')

# Test message
messages = ["Congratulations! You've won a free prize. Click here!"]
X_test = vectorizer.transform(messages)
prediction = model.predict(X_test)

print("SPAM" if prediction[0] == 1 else "NORMAL")