📧 Spam Message Classifier (Machine Learning Project)

An end-to-end Machine Learning project that classifies messages as Spam 🚨 or Not Spam ✅ using Natural Language Processing (NLP) techniques.

This project demonstrates the complete ML pipeline: data cleaning, feature engineering, model training, evaluation, and real-time prediction.

---

🚀 Project Overview

Spam messages are a common problem in emails and SMS. This project builds a machine learning model that can automatically detect whether a message is spam or not.

The model is trained on a large dataset of labeled messages and uses TF-IDF vectorization + classification algorithms to make predictions.

---

📂 Dataset

• Source: Kaggle Spam Dataset

• Total samples: 83,448 messages

• Features:

  • label → 0 (Not Spam), 1 (Spam)
  • text → Message content

---

🧠 Machine Learning Pipeline

1️⃣ Data Cleaning

• Removed missing values (none found)

• Filtered valid labels (0 and 1)

• Text preprocessing:

  • Lowercasing
  • Removing special characters
  • Removing extra spaces
  • Handling line-break noise (\n)

---

2️⃣ Feature Engineering

• Converted text into numerical features using:

  • TF-IDF Vectorization

• Improvements applied:

  • Stopword removal (stop_words='english')
  • N-grams (1 to 3 words) for better context understanding
  • Increased feature size for richer representation

---

3️⃣ Model Training

• Models explored:

  • Logistic Regression (with class balancing)
  • Multinomial Naive Bayes (optimized for text classification)

• Final model trained on processed data

---

4️⃣ Model Evaluation

📊 Performance Metrics:

• Accuracy: 98.24%
• Precision: 97.99%
• Recall: 98.67%
• F1 Score: 98.33%

📉 Confusion Matrix:

[[7761  177]
 [ 116 8636]]

🔍 Interpretation:

• High precision → Few false positives
• High recall → Most spam messages correctly detected
• Balanced performance across all metrics

---

🧪 Real-Time Prediction (User Input)

The model allows live testing through user input:

Example:

Enter a message: get this free offer today
Prediction: 🚨 Spam

Enter a message: hey are you coming to class?
Prediction: ✅ Not Spam

---

⚠️ Observations & Learnings

• The model sometimes misclassifies similar phrases:

  • "free offer" → Spam ✅
  • "free deal" → Not Spam ❌

Reason:

• Machine learning models learn from data patterns, not actual language meaning
• Some words (like "deal") may appear in both spam and normal messages

---

🔧 Improvements Applied

To enhance performance, the following optimizations were implemented:

• ✅ Stopword removal
• ✅ N-gram feature expansion (1–3 words)
• ✅ Class imbalance handling (class_weight='balanced')
• ✅ Tried Naive Bayes (better for text classification)
• ✅ Cleaned noisy text (line breaks, formatting issues)

---

🏁 Limitations

• Model does not fully understand semantic meaning
• Some edge cases still misclassified
• Performance depends heavily on dataset quality

---

💡 Future Improvements

• Use advanced NLP models (like word embeddings or transformers)
• Hyperparameter tuning
• Try Deep Learning models (LSTM, BERT)
• Deploy as a web app

---

🛠️ Tech Stack

• Python 🐍
• Pandas
• NumPy
• Scikit-learn
• NLP (TF-IDF)

---

📁 Project Structure

spam_classifier/
│── spam_classifier.py
│── combined_data.csv
│── README.md

---

▶️ How to Run

1. Clone the repository:

git clone <your-repo-link>

2. Install dependencies:

pip install pandas numpy scikit-learn

3. Run the script:

python spam_classifier.py

---

🎯 Key Takeaway

This project is not just about building a model, but understanding:

• How ML models behave
• Why predictions can go wrong
• How preprocessing impacts performance

---

⭐ Final Note

Even high-performing models are not perfect. The goal is to continuously improve, analyze mistakes, and learn from them.

---

🚀 This project marks a strong foundation in Machine Learning and NLP.