Project 4: Spam Email Detection

Classify emails as spam or not spam using text features.

Difficulty

Beginner

Time Estimate

2-3 days

Skills You'll Practice

• Text Classification
• NLP Basics
• Feature Engineering from Text
• Text Preprocessing

Learning Objectives

By completing this project, you will learn to:

• Preprocess text data
• Extract features from text
• Apply classification to text data
• Handle imbalanced text datasets
• Evaluate text classification models

Dataset

Option 1: SMS Spam Collection Dataset

• Kaggle SMS Spam Collection
• Simple CSV format
• Two columns: label (ham/spam) and text

Option 2: Email Spam Dataset

• Enron Spam Dataset
• More complex, real-world emails

Project Steps

Step 1: Load and Explore Data

• Load the dataset
• Check class distribution (spam vs ham)
• Analyze text length statistics
• Visualize word frequencies
• Check for missing values

Step 2: Text Preprocessing

• Convert to lowercase
• Remove punctuation
• Remove stopwords
• Tokenize text
• Stemming or lemmatization
• Remove special characters and numbers

Step 3: Feature Extraction

• Bag of Words (CountVectorizer)
• TF-IDF (TfidfVectorizer)
• N-grams (unigrams, bigrams)
• Text length features
• Word count features

Step 4: Model Training

• Split data into train/test sets
• Train multiple models:
  • Naive Bayes (great for text!)
  • Logistic Regression
  • Random Forest
  • SVM
• Compare model performance

Step 5: Model Evaluation

• Calculate accuracy, precision, recall, F1-score
• Create confusion matrix
• Plot ROC curve
• Analyze misclassified examples

Step 6: Model Improvement

• Tune hyperparameters
• Try different feature extraction methods
• Handle class imbalance
• Final model selection

Expected Deliverables

1. Jupyter Notebook with complete analysis:

   • Text preprocessing pipeline
   • Feature extraction
   • Model training and evaluation
   • Results and conclusions

2. Python Script (optional):

   • Function to predict spam/ham for new emails

Evaluation Metrics

• Accuracy: Overall correctness
• Precision: Of predicted spam, how many are actually spam
• Recall: Of actual spam, how many were detected
• F1-Score: Balance between precision and recall

Text Preprocessing Pipeline

# Example preprocessing steps
1. Lowercase conversion
2. Remove URLs, emails, phone numbers
3. Remove punctuation
4. Remove stopwords
5. Tokenization
6. Stemming/Lemmatization

Feature Extraction Methods

• Count Vectorizer: Word frequency
• TF-IDF: Term frequency-inverse document frequency
• N-grams: Word pairs, triplets
• Character n-grams: For catching typos

Tips

• Naive Bayes works very well for text classification
• TF-IDF usually performs better than simple word counts
• Try different n-gram ranges (1-2, 1-3)
• Visualize most common words in spam vs ham
• Handle class imbalance if needed
• Test on your own emails!

Resources

• Kaggle SMS Spam Dataset
• Scikit-learn Text Feature Extraction
• NLTK Documentation

Next Steps

After completing this project:

• Try more advanced NLP techniques
• Experiment with word embeddings
• Move to Project 5: Wine Quality