🛡️ ScamShield

AI-powered scam detection for SMS messages, URLs, phone numbers, emails, and files.

ScamShield uses a four-layer hybrid detection engine to identify scams in real time — combining rule-based pattern matching, a blacklist database, a machine learning model trained on 5,574 real SMS messages, and Azure AI Language sentiment analysis.

Live Demo →

---

Screenshots

Scanner with 5 scan types including file upload

Email scam detected — 87/100 risk score

Live statistics dashboard

PDF file upload and scanning

---

Why ScamShield

Scam messages targeting students and newcomers are increasingly sophisticated. Existing tools require apps or subscriptions. ScamShield is free, instant, and works on any device — paste a message, URL, email, or upload a file and get a risk score in under 200ms.

---

Features

• Four-layer detection — rules + blacklist + ML + Azure AI, combined into a single risk score (0–100)
• Five scan types — SMS/text, URL, phone number, email, and file upload (.txt, .eml, .csv, .msg, .pdf)
• PDF support — upload and scan PDF documents for scam content
• Scam explanations — human-readable explanation of why content was flagged
• Auto-detection — automatically identifies whether input is a message, URL, or phone number
• Scan history — stores and retrieves the last 10 scans
• Live statistics — tracks total scans, safe/suspicious/dangerous breakdown
• 97.2% ML accuracy — model trained on the UCI SMS Spam Collection dataset (5,574 messages)
• Azure AI Language — Layer 4 uses sentiment analysis to detect threatening tone

---

Architecture

User Input (text / URL / phone / email / file)
    │
    ▼
┌──────────────────────────────────────────────┐
│              FastAPI Backend                 │
│                                              │
│  ┌──────────┐  ┌────────────┐               │
│  │  Rule    │  │ Blacklist  │               │
│  │ Engine   │  │ (MongoDB)  │               │
│  │ (regex)  │  │            │               │
│  └────┬─────┘  └─────┬──────┘               │
│       │               │                      │
│       ▼               ▼                      │
│  ┌──────────────────────────────────────┐   │
│  │        ML Model (scikit-learn)       │   │
│  │   TF-IDF + Logistic Regression       │   │
│  │   Trained: 5,574 SMS messages        │   │
│  └──────────────────┬───────────────────┘   │
│                     │                        │
│                     ▼                        │
│  ┌──────────────────────────────────────┐   │
│  │        Azure AI Language             │   │
│  │   Sentiment Analysis (Layer 4)       │   │
│  │   Detects threatening tone & intent  │   │
│  └──────────────────┬───────────────────┘   │
│                     │                        │
│              Risk Score (0–100)              │
└──────────────────────────────────────────────┘
    │
    ▼
React Frontend (Vercel)

Detection layers:

Layer	Method	Max Score
Rule Engine	Regex patterns across 7 scam categories	70
Blacklist	MongoDB lookup of known scam domains/numbers/phrases	50
ML Layer	TF-IDF vectorizer + Logistic Regression	40
Azure AI	Sentiment analysis — threatening tone detection	30

Final score = sum of all layers, capped at 100.

---

Tech Stack

Layer	Technology
Backend	FastAPI, Python 3.11
Database	MongoDB Atlas (Motor async driver)
ML	scikit-learn (TF-IDF + Logistic Regression)
Cloud AI	Azure AI Language (sentiment analysis)
File Processing	PyPDF2 (PDF text extraction)
Frontend	React, React Router
Backend Deploy	Render
Frontend Deploy	Vercel

---

Local Setup

Backend

cd backend
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate
pip install -r requirements.txt

Create backend/.env:

MONGO_URL=your_mongodb_connection_string
DB_NAME=scamshield
AZURE_LANGUAGE_KEY=your_azure_key
AZURE_LANGUAGE_ENDPOINT=https://your-resource.cognitiveservices.azure.com/

uvicorn server:app --reload

Frontend

cd frontend
npm install

Create frontend/.env:

REACT_APP_API_BASE_URL=http://localhost:8000/api

npm start

---

API Endpoints

Method	Endpoint	Description
POST	/api/scan	Scan text, URL, phone, or email content
POST	/api/scan/file	Scan an uploaded file (.txt, .eml, .csv, .msg, .pdf)
GET	/api/history	Retrieve last 10 scan results
GET	/api/stats	Get aggregate scan statistics
GET	/api/health	Health check + model and Azure status

Example request

curl -X POST https://scamshield-fi4v.onrender.com/api/scan \
  -H "Content-Type: application/json" \
  -d '{"content": "URGENT: Your account will be suspended in 24 hours!"}'

Example response

{
  "id": "abc123",
  "content": "URGENT: Your account will be suspended in 24 hours!",
  "scan_type": "text",
  "risk_score": 85,
  "label": "🔴 Dangerous",
  "guidance": "This content is highly likely to be a scam.",
  "explanation": "Uses urgency tactics to pressure you into acting without thinking.",
  "triggers": ["Rule: urgency", "AI: suspicious_language_patterns", "Azure: high_negative_sentiment"],
  "timestamp": "2026-01-01T00:00:00Z"
}

---

ML Model Details

• Dataset: UCI SMS Spam Collection (5,574 messages — 747 spam, 4,827 ham)
• Vectorizer: TF-IDF (3,000 features, English stopwords removed)
• Classifier: Logistic Regression
• Train/test split: 80/20
• Test accuracy: 97.22%
• Model persistence: Saved to disk on first run, loaded on subsequent restarts

---

Azure AI Integration

ScamShield uses Azure AI Language as a fourth detection layer:

• Service: Azure AI Language (Text Analytics)
• Feature: Sentiment Analysis with opinion mining
• Logic: High negative sentiment (>70%) adds up to 30 points to the risk score
• Pricing tier: F0 (Free — 5,000 transactions/month)

This integration demonstrates Microsoft Azure ecosystem experience and adds a cloud AI signal on top of the local ML model.

---

Author

Ayesha Habib — University of Manitoba
GitHub | Live Demo