Grand Line Message Bounty Detector — NLP Classifier

Unsolicited promotional and fraudulent spam messages clutter user communication and pose security risks. Building a lightweight, high-precision NLP engine is essential to eliminate false alarms where critical messages are misclassified.

Scikit-learn unified pipeline encapsulating text preprocessing, sublinear TF-IDF vectorization (`sublinear_tf=True`), and a Multinomial Naive Bayes classifier. Evaluated on 5,572 records with stratified 80/20 train/test splitting.

• Raw dataset parsing and validation over 5,572 records (4,825 ham, 747 spam).
• Accent normalization and sublinear TF-IDF feature extraction inside pipeline.
• Stratified 80/20 train/test split ensuring zero data leakage.

• Multinomial Naive Bayes probabilistic intent scoring.
• Model serialization using `joblib` for instant real-time CLI inference.

Achieved 95.96% accuracy and 100.00% precision on evaluated test set (0 false positives), ensuring legitimate messages are never lost.

• Expand evaluation to MIME email header datasets (Enron Spam & SpamAssassin).
• Incorporate bi-gram and tri-gram n-gram features.
• Build a FastAPI microservice endpoint for live web UI integration.