Ameey-code/scamshield-nigeria-3mtt

# ScamShield Nigeria (FraudGuard AI) A machine learning-powered web app that detects fraudulent SMS messages, fake bank alerts, BVN/NIN threats, and scam job/investment offers targeting Nigerians. Built with Python, scikit-learn, and Streamlit. ## Problem Nigerians receive scam SMS messages daily — fake bank credit alerts, BVN/NIN restriction threats, and job/investment scams. Most people have no quick way to verify whether a message is genuinely risky before responding, calling, or clicking a link. ScamShield Nigeria gives an instant, evidence-based risk assessment for any pasted message. ## Features - **Fraud Risk Scoring** — every message gets a 0–100% scam probability, color-coded (green/amber/red) - **Known-Scammer Number Lookup** — messages containing a previously reported phone number are instantly flagged, independent of the ML model's judgment - **Explainable AI** — shows the actual words the trained model weighted most heavily toward "scam," not just a fixed keyword list - **Scam-Type Pattern Matching** — identifies which known Nigerian scam category a message resembles (BVN/NIN threat, fake credit alert, job scam, lottery scam) or confirms a legitimate transaction pattern - **Red-Flag Keyword Extraction** — highlights specific warning signs (urgency language, phone numbers, links) - **Bulk Scan** — paste multiple messages at once (e.g. a forwarded chain) and get a scored table - **Context-Aware Safety Recommendations** — different guidance depending on risk level - **System Insights Dashboard** — live model accuracy, precision, recall, and dataset breakdown ## Tech Stack - Python 3.10+ - scikit-learn (Logistic Regression classifier, TF-IDF vectorization) - Streamlit (web interface) - pandas, numpy, nltk, joblib ## Project Structure fraud_sms_app/ ├── data/ │ ├── spam_raw.csv # Kaggle SMS Spam Collection dataset │ ├── spam_nigerian_augmented.csv # Combined + cleaned training data │ └── known_scammers.csv # Locally reported scam numbers ├── models/ …