📧 Email Spam Detection with Multinomial Naive Bayes Classifier

Table of Contents

1. Project Overview
2. Algorithm Used (Multinomial Naive Bayes)
3. Dataset
4. Project Structure
5. Setup and Installation
6. How to Run the Streamlit App
7. Results and Visualization
8. Conclusion

---

1. Project Overview

This project implements a robust Machine Learning solution to classify incoming messages as either "spam" or "ham" (not spam). The goal is to build an effective filter based on the text content of the email/SMS. We utilize a highly efficient classification algorithm and deploy the final model using an interactive Streamlit web application.

Key Features:

• Data Preprocessing: Cleaning and tokenization of text data.
• Feature Extraction: Using TF-IDF (Term Frequency-Inverse Document Frequency) vectorization.
• Model Training: Training a Multinomial Naive Bayes classifier.
• Interactive App: A Streamlit interface to test custom text for spam probability in real-time.

---

2. Algorithm Used (Multinomial Naive Bayes)

The core classification is performed using the Multinomial Naive Bayes algorithm.

• How it Works: This model is one of the standard algorithms for text classification. It's based on Bayes' theorem and is well-suited for discrete feature counts (like word counts or TF-IDF values). It calculates the probability of a message belonging to a class (spam or ham) given the frequency of its words.
• Suitability for Text: The Multinomial distribution models the probability of observing word counts in a document given the class, making it exceptionally effective for high-dimensional, sparse data like text feature vectors.
• Feature Vectorization: Text is transformed using TF-IDF (Term Frequency-Inverse Document Frequency), which assigns weights to words based on their importance. This helps the Naive Bayes model identify words that are highly characteristic of spam (e.g., "free," "win," "click").

---

3. Dataset

The project uses a publicly available SMS Spam Collection Dataset (often used for this task).

Column	Description	Example
target	The class label (0 for ham, 1 for spam)	spam or ham
text	The raw message/email content	"Call now to win a free prize!"

Note: The dataset requires cleaning to remove punctuation, stop words, and perform stemming/lemmatization.

---

4. Project Structure

The repository is organized as follows: email-spam-detection/ ├── .gitignore ├── README.md ├── requirements.txt # Lists all necessary Python libraries ├── spam_classifier.py # Main ML code: loads data, cleans, trains Naive Bayes, saves model and vectorizer. └── streamlit_app.py # Streamlit code for the interactive web interface.

---

5. Setup and Installation

To run this project locally, follow these steps:

1. Clone the repository:

   git clone [https://github.com/YourUsername/email-spam-detection.git](https://github.com/YourUsername/email-spam-detection.git)
   cd email-spam-detection

2. Create and activate a virtual environment (Recommended):

   python -m venv venv
   source venv/bin/activate  # On Windows, use `venv\Scripts\activate`

3. Install the required libraries:

   pip install -r requirements.txt

4. Run the ML Training Script: This trains the Multinomial Naive Bayes model and saves both the model and the TF-IDF vectorizer (which is needed by the Streamlit app for prediction).

   python spam_classifier.py

---

6. How to Run the Streamlit App

The project includes an interactive Streamlit application for demonstration.

1. Ensure you have completed the Setup steps above.

2. Run the Streamlit app:

   streamlit run streamlit_app.py

3. The app will automatically open in your web browser at a local address (usually http://localhost:8501).

📷 Streamlit Application Preview

Figure 1: Streamlit App Interface showing the text input box and the classification result (Spam/Ham).

---

7. Results and Visualization

Model Performance

The Multinomial Naive Bayes classifier performed well, achieving the following metrics on the test set:

• Accuracy: [Insert your calculated Accuracy Score]%
• Precision (Spam): [Insert Spam Precision Score]% (Crucial metric for spam detection to minimize false positives, i.e., marking a legitimate email as spam.)

Key Visualization

A visualization of word clouds can highlight the difference in vocabulary between the two classes, justifying the model's success.

Figure 2: Word Cloud visualizing the most frequent, high-weighted words found in 'Spam' messages, demonstrating distinct vocabulary usage and feature separation.

---

8. Conclusion

The combination of TF-IDF for feature extraction and the Multinomial Naive Bayes classifier provided a fast and highly accurate solution for email/SMS spam detection. The resulting model is robust and suitable for immediate deployment, as showcased by the interactive Streamlit application.