📧 Email/SMS Spam Classifier

This is a machine learning project that classifies emails and SMS messages as spam or not spam using Natural Language Processing (NLP) where I have used Navie Based ( Multinomial Naive Bayes ) which assumes that all the word in a sentence does not have correlation, this may not be true in practical situations but when I tried this model it gave some really good scores of 97.10 % accuracy and a precison of 100 % after using several EDA ( Exploratory Data Analysis ) methods and feature engineering methods I broke the dataset into 80 % training and 20 % tesitng and then I was able to get usefull features to feed the model.

Second Enhancement

Further I used docker to containarize my model so that anyone can use docker image and build my project locally and don't have to install all the dependencies. I also hosted this model on Huggingface which provided me smooth integration of model backend with streamlit frontend.

👨‍💻 Author

Shrish Mishra

🎯 Project Overview

This project implements a spam detection system using multiple machine learning algorithms. The system processes text messages, transforms them using TF-IDF vectorization, and classifies them as spam or legitimate (ham) messages.

📚 Dataset

• Source: SMS Spam Collection Dataset
• Total Messages: 5,572 messages
• After Preprocessing: 5,169 messages (after removing 403 duplicates)
• Distribution:
  • Ham (Not Spam): 4,516 messages (87.37%)
  • Spam: 653 messages (12.63%)

⚙️ Features

The project analyzes the following text features:

• Number of characters
• Number of words
• Number of sentences
• Transformed text (after preprocessing)

🔧 Text Preprocessing Pipeline

The transform_text function in Spam_detection.ipynb and app.py performs the following steps:

1. Lowercase Conversion: Converts all text to lowercase
2. Tokenization: Breaks text into individual words using nltk.word_tokenize()
3. Alphanumeric Filtering: Removes special characters and keeps only alphanumeric tokens
4. Stop Words Removal: Removes common English stop words and punctuation
5. Stemming: Reduces words to their root form using Porter Stemmer

🔢 Vectorization

TF-IDF Vectorizer (Term Frequency-Inverse Document Frequency)

• Maximum features: 3,000
• Converts text into numerical vectors
• Saved as: vectorizer.pkl

🤖 Machine Learning Models

The project evaluates 10 different classification algorithms:

📊 Model Performance Comparison

Algorithm	Accuracy	Precision
Multinomial Naive Bayes (NB)	97.10%	100.00%
K-Nearest Neighbors (KN)	90.52%	100.00%
Random Forest (RF)	97.58%	98.29%
Support Vector Classifier (SVC)	97.58%	97.48%
Extra Trees Classifier (ETC)	97.49%	97.46%
Logistic Regression (LR)	95.84%	97.03%
Gradient Boosting (GBDT)	94.68%	91.92%
Bagging Classifier (BgC)	95.84%	86.82%
AdaBoost	92.46%	84.88%
Decision Tree (DT)	92.75%	81.19%

🏆 Best Performing Model

Multinomial Naive Bayes was selected as the final model due to:

• High accuracy: 97.10%
• Perfect precision: 100.00%
• No false positives (0 legitimate messages classified as spam)
• Confusion Matrix:

  [[896   0]
   [ 30 108]]
  

• Model saved as: model.pkl 💾

📁 Project Structure

spam_detection_model/
├── app.py                      # Streamlit web application 
├── Spam_detection.ipynb        # Jupyter notebook with full analysis 
├── spam.csv                    # Dataset 
├── model.pkl                   # Trained Multinomial Naive Bayes model 
└── vectorizer.pkl             # TF-IDF vectorizer 

🛠️ Technologies Used

• Python 3.9.6 🐍
• Libraries:
  • pandas
  • numpy
  • nltk
  • scikit-learn
  • matplotlib
  • seaborn
  • wordcloud
  • streamlit
  • pickle

💻 Installation

1. Clone the repository
2. Install required packages:

pip install pandas numpy nltk scikit-learn matplotlib seaborn wordcloud streamlit

3. Download NLTK data: ⬇️

import nltk
nltk.download('stopwords')
nltk.download('punkt')
nltk.download('punkt_tab')

🚀 Usage

🌐 Running the Web Application

streamlit run app.py

The application provides a simple interface where you can:

1. Enter an email or SMS message
2. Click "Predict"
3. See if the message is classified as "Spam" or "Not Spam"

📓 Using the Notebook

Open Spam_detection.ipynb in Jupyter Notebook to:

• Explore the complete data analysis
• View visualizations (word clouds, histograms, correlation heatmaps)
• Train and evaluate different models
• Modify and experiment with the code

🎓 Model Training Process

1. Data Loading: Read spam.csv with ISO-8859-1 encoding
2. Data Cleaning:
   • Remove unnecessary columns
   • Rename columns to 'target' and 'text'
   • Encode labels (ham=0, spam=1)
   • Remove 403 duplicate messages
3. Feature Engineering: Extract character count, word count, and sentence count
4. Text Preprocessing: Apply the transform_text function
5. Vectorization: Convert text to TF-IDF features (3000 features)
6. Train-Test Split: 80% training, 20% testing (random_state=2)
7. Model Training: Train 10 different classifiers
8. Evaluation: Compare accuracy and precision scores
9. Model Selection: Choose Multinomial Naive Bayes as the final model

💡 Key Insights

• Spam messages are significantly longer than ham messages 📏
• Average characters:
  • Ham: 70.46
  • Spam: 137.89
• Average words:
  • Ham: 17.12
  • Spam: 27.67
• Most frequent words in spam: "call", "free", "txt", "claim", "prize"

🌐 Deployment

Run locally with:

streamlit run app.py

Docker (Build and Run)

You can run the app inside Docker — this is useful for consistent environments and for deploying to servers or CI.

Build the Docker image locally (from the repository root):

docker build -t spam-detection-app:latest .

Run the container and map Streamlit's port 8501 to your host:

docker run --rm -p 8501:8501 -v "$PWD":/app spam-detection-app:latest

Or use docker-compose (recommended for development):

docker compose up --build

Example: run the container and pass a private HF token (if needed):

docker run --rm -p 8501:8501 -e HF_TOKEN="$HF_TOKEN" -v "$PWD":/app spam-detection-app:latest

🚀 Future Improvements

• Add more sophisticated preprocessing techniques
• Implement deep learning models (LSTM, BERT)
• Add multilingual support
• Enhance the web interface with more features
• Add confidence scores and probability display
• Implement feedback mechanism for model improvement

📄 License

This project is open source and available for educational purposes. 📖

📧 Contact

For questions or suggestions, please contact Shrish Mishra. 💬

Future Improvements

• Add more sophisticated preprocessing techniques
• Implement deep learning models (LSTM, BERT)
• Add multilingual support
• Enhance the web interface with more features
• Deploy to cloud platform (Heroku, AWS, etc.)

Contact

For questions or suggestions, please contact at shrish409@gmail.com