📧 Spam Email Detection using Classical Machine Learning

A complete end-to-end Natural Language Processing (NLP) project that detects spam SMS messages using classical machine learning algorithms.

The project explores the complete NLP workflow:

• 📊 Exploratory Data Analysis
• 🧹 Text Preprocessing
• 🔤 TF-IDF Feature Extraction
• 🤖 Multiple Machine Learning Models
• 📈 Model Benchmarking
• 🔍 Error Analysis
• 💾 Model Serialization

---

📌 Project Workflow

SMS Dataset
      │
      ▼
Data Cleaning
      │
      ▼
Exploratory Data Analysis
      │
      ▼
Text Preprocessing

• Lowercase
• Tokenization
• Stopword Removal
• Punctuation Removal
• Stemming

      │
      ▼
Feature Engineering

• Character Count
• Word Count
• Sentence Count

      │
      ▼
TF-IDF Vectorization

      │
      ▼
Train/Test Split

      │
      ▼
Model Benchmark

├── Naive Bayes
├── Logistic Regression
├── Linear SVM
├── Random Forest
├── Gradient Boosting

      │
      ▼
Evaluation

Accuracy
Precision
Recall
F1 Score
ROC Curve
Confusion Matrix

      │
      ▼
Best Model

Linear SVM

      │
      ▼
Prediction System

      │
      ▼
Model Serialization

---

📊 Dataset

SMS Spam Collection Dataset

• 5,169 SMS messages (after cleaning)
• Binary Classification

Classes:

• ✅ Ham
• 🚨 Spam

---

📚 Technologies Used

• Python
• Pandas
• NumPy
• Matplotlib
• Seaborn
• NLTK
• Scikit-Learn
• Joblib

---

📂 Repository Structure

spam-email-detection/
│
├── notebook/
│     spam-message-detector.ipynb
│
├── images/
│     class_distribution.png
│     spam_wordcloud.png
│     ham_wordcloud.png
│     spam_words.png
│     ham_words.png
│     confusion_matrix.png
│     roc_curve.png
│     feature_importance.png
│
├── models/
│     linear_svm_spam_classifier.pkl
│     tfidf_vectorizer.pkl
│
├── requirements.txt
├── LICENSE
└── README.md

---

📸 Project Visualizations

📊 Class Distribution

---

☁️ Spam Word Cloud

---

☁️ Ham Word Cloud

---

📈 Most Frequent Spam Words

---

📉 Most Frequent Ham Words

---

🔲 Confusion Matrix

---

📈 ROC Curve

---

🔍 Most Important Spam Features

---

🏆 Model Benchmark

Model	Accuracy	Precision	Recall	F1 Score
Linear SVM	97.97%	97.41%	86.26%	91.50%
Random Forest	97.58%	99.07%	81.68%	89.54%
Naive Bayes	97.00%	98.08%	77.86%	86.81%
Logistic Regression	96.23%	100%	70.22%	82.51%
Gradient Boosting	96.03%	95.91%	71.76%	82.10%

---

🎯 Key Findings

• TF-IDF effectively transformed textual information into numerical representations suitable for machine learning.
• Linear SVM achieved the best balance between precision and recall.
• Spam messages generally contain:
  • More words
  • More characters
  • Promotional keywords
• Classical Machine Learning remains highly effective for spam detection.

---

🚀 Prediction System

Example:

predict_spam(
    "Congratulations! You've won a FREE iPhone!"
)

Output

🚨 Spam

---

💾 Saved Models

The repository contains:

• Linear SVM Classifier
• TF-IDF Vectorizer

Both are serialized using Joblib and can be directly loaded for deployment.

---

⚙️ Installation

Clone the repository

git clone https://github.com/yourusername/spam-email-detection.git

cd spam-email-detection

Install dependencies

pip install -r requirements.txt

---

▶️ Running the Notebook

jupyter notebook

Open

notebook/spam-message-detector.ipynb

Run all cells.

---

🚀 Future Improvements

• Hyperparameter Optimization
• Bi-gram & Tri-gram TF-IDF
• Lemmatization
• LSTM Spam Detection
• BERT-based Spam Classification
• Streamlit Deployment
• REST API using FastAPI

---

⭐ If you found this project useful

Please consider giving the repository a ⭐.