Upload complete fraud API project

DEPLOYMENT.md

@@ -0,0 +1,179 @@
+# Deployment Guide - HuggingFace Spaces
+
+## Prerequisites
+
+- HuggingFace account
+- Git installed locally
+- Trained model files
+
+## Step-by-Step Deployment
+
+### 1. Create a New Space
+
+1. Go to https://huggingface.co/new-space
+2. Choose a name for your space (e.g., `fraud-detection-api`)
+3. Select **Docker** as the SDK
+4. Choose visibility (Public or Private)
+5. Click "Create Space"
+
+### 2. Initialize Git Repository
+
+```bash
+cd fraud_api
+git init
+git add .
+git commit -m "Initial commit: Fraud Detection API"
+```
+
+### 3. Add HuggingFace Remote
+
+```bash
+# Replace YOUR_USERNAME and YOUR_SPACE with your details
+git remote add origin https://huggingface.co/spaces/YOUR_USERNAME/YOUR_SPACE
+```
+
+### 4. Push to HuggingFace
+
+```bash
+git push -u origin main
+```
+
+**Note:** You may be prompted for credentials:
+- Username: Your HuggingFace username
+- Password: Use a **HuggingFace Access Token** (not your password)
+  - Get token from: https://huggingface.co/settings/tokens
+
+### 5. Monitor Build
+
+1. Go to your Space URL: `https://huggingface.co/spaces/YOUR_USERNAME/YOUR_SPACE`
+2. Click on "Logs" tab to monitor the Docker build
+3. Build typically takes 3-5 minutes
+
+### 6. Access Your API
+
+Once deployed, your API will be available at:
+```
+https://YOUR_USERNAME-YOUR_SPACE.hf.space
+```
+
+Test it:
+```bash
+curl https://YOUR_USERNAME-YOUR_SPACE.hf.space/health
+```
+
+## Troubleshooting
+
+### Build Fails - Missing Models
+
+**Problem:** Models not found in `models/` directory
+
+**Solution:**
+1. Ensure model files are committed to git
+2. Check `.gitignore` doesn't exclude `.joblib` files
+3. Verify models are in correct location
+
+### Out of Memory Error
+
+**Problem:** Docker container runs out of memory
+
+**Solution:**
+1. Reduce model size (use only necessary models)
+2. Implement lazy loading
+3. Request more resources from HuggingFace
+
+### Port Issues
+
+**Problem:** Application not accessible
+
+**Solution:**
+Ensure Dockerfile uses port 7860 (HuggingFace standard):
+```dockerfile
+EXPOSE 7860
+CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860"]
+```
+
+## Updating Your Deployment
+
+When you make changes:
+
+```bash
+git add .
+git commit -m "Update: description of changes"
+git push origin main
+```
+
+HuggingFace will automatically rebuild and redeploy.
+
+## Advanced Configuration
+
+### Environment Variables
+
+Add secrets via HuggingFace Space settings:
+1. Go to Space Settings → Repository secrets
+2. Add key-value pairs
+3. Access in `app.py`:
+
+```python
+import os
+SECRET_KEY = os.getenv("SECRET_KEY")
+```
+
+### Custom Domain
+
+For production, consider:
+1. Upgrading to HuggingFace Pro
+2. Setting up custom domain
+3. Adding CDN/caching layer
+
+## Monitoring
+
+### Check Logs
+
+```bash
+# View real-time logs in HuggingFace UI
+# Or use API:
+curl https://huggingface.co/api/spaces/YOUR_USERNAME/YOUR_SPACE/logs
+```
+
+### Usage Analytics
+
+HuggingFace provides basic analytics:
+- Request count
+- Response times
+- Error rates
+
+Access from Space settings dashboard.
+
+## Cost Considerations
+
+**Free Tier:**
+- Limited CPU/RAM
+- May sleep after inactivity
+- Suitable for demos/testing
+
+**Paid Options:**
+- Persistent compute
+- GPU access
+- Higher resource limits
+- Custom containers
+
+## Security Checklist
+
+Before going to production:
+
+- [ ] Add authentication
+- [ ] Implement rate limiting
+- [ ] Set up CORS properly
+- [ ] Use HTTPS only
+- [ ] Monitor for abuse
+- [ ] Set resource limits
+- [ ] Add input validation
+- [ ] Implement logging
+- [ ] Regular security updates
+- [ ] Model versioning strategy
+
+## Support
+
+- Documentation: https://huggingface.co/docs/hub/spaces-overview
+- Community: https://discuss.huggingface.co
+- Issues: https://github.com/huggingface/hub-docs/issues

Dockerfile

@@ -0,0 +1,22 @@
+FROM python:3.10-slim
+
+WORKDIR /code
+
+# Install system dependencies
+RUN apt-get update && apt-get install -y \
+    gcc \
+    g++ \
+    && rm -rf /var/lib/apt/lists/*
+
+# Copy requirements and install Python dependencies
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+
+# Copy application code
+COPY . .
+
+# Expose port for HuggingFace Spaces
+EXPOSE 7860
+
+# Run the application
+CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860"]

README.md

@@ -1,12 +1,233 @@
----
-title: Fraud Detector
-emoji: 📊
-colorFrom: blue
-colorTo: red
-sdk: docker
-pinned: false
-license: other
-short_description: 'Insurance Fraud Detection '
----
-
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# Fraud Detection API
+
+Production-ready inference API for insurance fraud detection using pre-trained ML models.
+
+## 🚀 Quick Start
+
+### Local Development
+
+1. **Copy your trained models:**
+   ```bash
+   cp ../models/best_tree_models_calibrated.joblib models/
+   cp ../models/best_tree_models_uncalibrated.joblib models/
+   ```
+
+2. **Install dependencies:**
+   ```bash
+   pip install -r requirements.txt
+   ```
+
+3. **Run the server:**
+   ```bash
+   uvicorn app:app --reload --port 7860
+   ```
+
+4. **Open the UI:**
+   Visit `http://localhost:7860` in your browser
+
+### Example API Request
+
+```bash
+curl -X POST "http://localhost:7860/predict?model=xgb&scenario=dashboard" \
+  -H "Content-Type: application/json" \
+  -d '{
+    "policy_annual_premium": 1200.0,
+    "total_claim_amount": 15000.0,
+    "vehicle_age": 5,
+    "days_since_bind": 300,
+    "months_as_customer": 24,
+    "capital-gains": 0,
+    "capital-loss": 0,
+    "injury_share": 0.4,
+    "property_share": 0.6,
+    "umbrella_limit": 0,
+    "incident_hour_of_the_day": 14
+  }'
+```
+
+### Example Response
+
+```json
+{
+  "model": "XGBoost",
+  "calibrated": true,
+  "probability": 0.73,
+  "threshold_flag": null,
+  "scenario": "dashboard"
+}
+```
+
+## 📋 API Reference
+
+### Endpoints
+
+#### `POST /predict`
+
+Make a fraud prediction for an insurance claim.
+
+**Query Parameters:**
+- `model` (string): Model type - `rf` (RandomForest), `et` (ExtraTrees), or `xgb` (XGBoost)
+- `scenario` (string): `dashboard` (calibrated) or `auto_flagger` (uncalibrated + threshold)
+- `calibrated` (boolean): Override calibration (optional, scenario takes precedence)
+
+**Request Body:**
+```json
+{
+  "policy_annual_premium": float,
+  "total_claim_amount": float,
+  "vehicle_age": int,
+  "days_since_bind": int,
+  "months_as_customer": int,
+  "capital-gains": float,
+  "capital-loss": float,
+  "injury_share": float,
+  "property_share": float,
+  "umbrella_limit": int,
+  "incident_hour_of_the_day": int (0-23)
+}
+```
+
+#### `GET /health`
+
+Health check endpoint returning model status.
+
+## 🎯 Deployment Scenarios
+
+### Scenario A: Auto-Flagger
+**Use Case:** Automated claim flagging system
+
+- Uses **uncalibrated** models for maximum recall
+- Returns decision flag: `AUTO_FLAG` or `AUTO_APPROVE`
+- Threshold: 0.53 (adjust based on your F2 optimization)
+
+```bash
+curl -X POST "http://localhost:7860/predict?model=xgb&scenario=auto_flagger" \
+  -H "Content-Type: application/json" \
+  -d @claim_data.json
+```
+
+### Scenario B: Investigator Dashboard
+**Use Case:** Human-in-the-loop prioritization
+
+- Uses **calibrated** models for accurate probabilities
+- Returns probability score for ranking claims
+- No hard threshold decision
+
+```bash
+curl -X POST "http://localhost:7860/predict?model=xgb&scenario=dashboard" \
+  -H "Content-Type: application/json" \
+  -d @claim_data.json
+```
+
+## 🐳 Docker Deployment
+
+### Build and Run Locally
+
+```bash
+docker build -t fraud-api .
+docker run -p 7860:7860 fraud-api
+```
+
+### Deploy to HuggingFace Spaces
+
+1. Create a new Space on HuggingFace
+2. Select **Docker** as SDK
+3. Push this folder to your Space repository:
+
+```bash
+git init
+git add .
+git commit -m "Initial commit"
+git remote add origin https://huggingface.co/spaces/YOUR_USERNAME/YOUR_SPACE_NAME
+git push -u origin main
+```
+
+4. HuggingFace will automatically build and deploy your Docker container
+5. Your API will be available at: `https://YOUR_USERNAME-YOUR_SPACE_NAME.hf.space`
+
+## 📁 Project Structure
+
+```
+fraud_api/
+├── app.py              # FastAPI backend
+├── index.html          # Web UI
+├── requirements.txt    # Python dependencies
+├── Dockerfile          # Container configuration
+├── README.md           # This file
+└── models/             # Model files (add your .joblib files here)
+    ├── best_tree_models_calibrated.joblib
+    └── best_tree_models_uncalibrated.joblib
+```
+
+## ⚙️ Configuration
+
+### Adjust Auto-Flag Threshold
+
+Edit `app.py` line 19:
+```python
+THRESHOLD_AUTO_FLAG = 0.53  # Adjust based on your requirements
+```
+
+### Model Loading
+
+Models are loaded on startup from `models/` directory. Expected format:
+```python
+{
+  'Trees': {
+    'RandomForest': <model_pipeline>,
+    'ExtraTrees': <model_pipeline>,
+    'XGBoost': <model_pipeline>
+  }
+}
+```
+
+## 🛠️ Testing
+
+Test the API with sample data:
+
+```bash
+# High-risk claim
+curl -X POST "http://localhost:7860/predict?model=xgb&scenario=auto_flagger" \
+  -H "Content-Type: application/json" \
+  -d '{
+    "policy_annual_premium": 500,
+    "total_claim_amount": 50000,
+    "vehicle_age": 1,
+    "days_since_bind": 10,
+    "months_as_customer": 2,
+    "capital-gains": 10000,
+    "capital-loss": 0,
+    "injury_share": 0.8,
+    "property_share": 0.2,
+    "umbrella_limit": 0,
+    "incident_hour_of_the_day": 3
+  }'
+```
+
+## 📊 Model Information
+
+This API serves predictions from models trained on insurance claim data with F2-score optimization for fraud detection. The models were calibrated using Platt scaling to ensure probability quality.
+
+**Available Models:**
+- **RandomForest**: Ensemble of decision trees
+- **ExtraTrees**: Extra randomized trees
+- **XGBoost**: Gradient boosted decision trees
+
+**Calibration:**
+- Uncalibrated: Optimized for maximum recall (catching fraud)
+- Calibrated: Optimized for probability accuracy (ranking)
+
+## 🔒 Security Notes
+
+- This is a minimal inference API for demonstration
+- For production deployment, add:
+  - Authentication (API keys, OAuth)
+  - Rate limiting
+  - Input sanitization
+  - HTTPS/TLS
+  - Monitoring and logging
+  - Model versioning
+
+## 📝 License
+
+MIT License - See project root for details

app.py

@@ -0,0 +1,201 @@
+"""
+Fraud Detection API - FastAPI Backend
+Serves predictions from pre-trained ML models (RandomForest, ExtraTrees, XGBoost)
+Supports both calibrated and uncalibrated versions with two deployment scenarios.
+"""
+
+from fastapi import FastAPI, HTTPException, Query
+from fastapi.staticfiles import StaticFiles
+from fastapi.responses import FileResponse
+from pydantic import BaseModel, Field
+from typing import Optional, Literal
+import joblib
+import numpy as np
+from pathlib import Path
+import logging
+
+# Configure logging
+logging.basicConfig(level=logging.INFO)
+logger = logging.getLogger(__name__)
+
+# Initialize FastAPI app
+app = FastAPI(title="Fraud Detection API", version="1.0.0")
+
+# Model configuration
+MODELS_DIR = Path("models")
+THRESHOLD_AUTO_FLAG = 0.53  # Placeholder - adjust based on your F2 optimization
+
+# Model registry
+MODELS = {}
+
+class ClaimInput(BaseModel):
+    """Input schema for claim predictions"""
+    policy_annual_premium: float = Field(..., description="Annual policy premium")
+    total_claim_amount: float = Field(..., description="Total claim amount")
+    vehicle_age: int = Field(..., description="Age of vehicle in years")
+    days_since_bind: int = Field(..., description="Days since policy binding")
+    months_as_customer: int = Field(..., description="Months as customer")
+    capital_gains: float = Field(0.0, alias="capital-gains")
+    capital_loss: float = Field(0.0, alias="capital-loss")
+    injury_share: float = Field(..., description="Share of injury damage")
+    property_share: float = Field(..., description="Share of property damage")
+    umbrella_limit: int = Field(..., description="Umbrella policy limit")
+    incident_hour_of_the_day: int = Field(..., ge=0, le=23)
+    hour_sin: Optional[float] = None
+    hour_cos: Optional[float] = None
+    
+    class Config:
+        populate_by_name = True
+
+class PredictionResponse(BaseModel):
+    """Response schema for predictions"""
+    model: str
+    calibrated: bool
+    probability: float
+    threshold_flag: Optional[str] = None
+    scenario: str
+
+def load_models():
+    """Load all available models on startup"""
+    model_types = ["RandomForest", "ExtraTrees", "XGBoost"]
+    calibration_types = ["calibrated", "uncalibrated"]
+    
+    for model_type in model_types:
+        for cal_type in calibration_types:
+            # Expected filename format: best_tree_models_calibrated.joblib or best_tree_models_uncalibrated.joblib
+            filename = f"best_tree_models_{cal_type}.joblib"
+            filepath = MODELS_DIR / filename
+            
+            if filepath.exists():
+                try:
+                    models_dict = joblib.load(filepath)
+                    # Models are stored in dict structure: {'Trees': {'RandomForest': model, 'XGBoost': model, ...}}
+                    if 'Trees' in models_dict and model_type in models_dict['Trees']:
+                        key = f"{model_type}_{cal_type}"
+                        MODELS[key] = models_dict['Trees'][model_type]
+                        logger.info(f"Loaded model: {key}")
+                except Exception as e:
+                    logger.error(f"Error loading {filepath}: {e}")
+    
+    logger.info(f"Total models loaded: {len(MODELS)}")
+    if not MODELS:
+        logger.warning("No models loaded! Check models directory.")
+
+@app.on_event("startup")
+async def startup_event():
+    """Load models on application startup"""
+    load_models()
+
+@app.get("/")
+async def root():
+    """Serve the frontend HTML"""
+    return FileResponse("index.html")
+
+@app.get("/health")
+async def health_check():
+    """Health check endpoint"""
+    return {
+        "status": "healthy",
+        "models_loaded": len(MODELS),
+        "available_models": list(MODELS.keys())
+    }
+
+@app.post("/predict", response_model=PredictionResponse)
+async def predict(
+    claim_data: ClaimInput,
+    model: Literal["rf", "et", "xgb"] = Query("rf", description="Model type: rf=RandomForest, et=ExtraTrees, xgb=XGBoost"),
+    calibrated: bool = Query(True, description="Use calibrated model"),
+    scenario: Literal["auto_flagger", "dashboard"] = Query("dashboard", description="Prediction scenario")
+):
+    """
+    Predict fraud probability for an insurance claim.
+    
+    - **Scenario A (auto_flagger)**: Uses uncalibrated model + threshold for auto-flagging
+    - **Scenario B (dashboard)**: Uses calibrated model for ranking/prioritization
+    """
+    
+    # Map shorthand to full model names
+    model_map = {"rf": "RandomForest", "et": "ExtraTrees", "xgb": "XGBoost"}
+    model_name = model_map[model]
+    
+    # Determine calibration type
+    cal_type = "calibrated" if calibrated else "uncalibrated"
+    model_key = f"{model_name}_{cal_type}"
+    
+    # Override calibration based on scenario
+    if scenario == "auto_flagger":
+        cal_type = "uncalibrated"
+        model_key = f"{model_name}_uncalibrated"
+    elif scenario == "dashboard":
+        cal_type = "calibrated"
+        model_key = f"{model_name}_calibrated"
+    
+    # Get model
+    if model_key not in MODELS:
+        raise HTTPException(
+            status_code=404,
+            detail=f"Model {model_key} not found. Available: {list(MODELS.keys())}"
+        )
+    
+    loaded_model = MODELS[model_key]
+    
+    # Prepare input data
+    # Calculate hour_sin and hour_cos if not provided
+    if claim_data.hour_sin is None or claim_data.hour_cos is None:
+        hour_rad = (claim_data.incident_hour_of_the_day / 24) * 2 * np.pi
+        claim_data.hour_sin = np.sin(hour_rad)
+        claim_data.hour_cos = np.cos(hour_rad)
+    
+    # Convert to dict and create feature array
+    # Note: The model expects the preprocessor to handle feature engineering
+    # We'll pass raw features as a dict
+    features_dict = claim_data.dict(by_alias=True)
+    
+    # For deployment, you would typically have a preprocessor that was saved with the model
+    # Here we assume the model is already wrapped in a pipeline that handles preprocessing
+    try:
+        # Create input array - order must match training
+        # The pipeline should handle the transformation
+        input_data = {
+            'policy_annual_premium': features_dict['policy_annual_premium'],
+            'total_claim_amount': features_dict['total_claim_amount'],
+            'vehicle_age': features_dict['vehicle_age'],
+            'days_since_bind': features_dict['days_since_bind'],
+            'months_as_customer': features_dict['months_as_customer'],
+            'capital-gains': features_dict['capital-gains'],
+            'capital-loss': features_dict['capital-loss'],
+            'injury_share': features_dict['injury_share'],
+            'property_share': features_dict['property_share'],
+            'umbrella_limit': features_dict['umbrella_limit'],
+            'incident_hour_of_the_day': features_dict['incident_hour_of_the_day'],
+            'hour_sin': features_dict['hour_sin'],
+            'hour_cos': features_dict['hour_cos']
+        }
+        
+        # If model is a pipeline, it expects a DataFrame
+        import pandas as pd
+        input_df = pd.DataFrame([input_data])
+        
+        # Get prediction probability
+        proba = loaded_model.predict_proba(input_df)[0, 1]  # Probability of fraud (class 1)
+        
+    except Exception as e:
+        logger.error(f"Prediction error: {e}")
+        raise HTTPException(status_code=500, detail=f"Prediction failed: {str(e)}")
+    
+    # Determine threshold flag for auto_flagger scenario
+    threshold_flag = None
+    if scenario == "auto_flagger":
+        threshold_flag = "AUTO_FLAG" if proba >= THRESHOLD_AUTO_FLAG else "AUTO_APPROVE"
+    
+    return PredictionResponse(
+        model=model_name,
+        calibrated=(cal_type == "calibrated"),
+        probability=float(proba),
+        threshold_flag=threshold_flag,
+        scenario=scenario
+    )
+
+if __name__ == "__main__":
+    import uvicorn
+    uvicorn.run(app, host="0.0.0.0", port=7860)

example_claim.json

@@ -0,0 +1,13 @@
+{
+    "policy_annual_premium": 1200.0,
+    "total_claim_amount": 15000.0,
+    "vehicle_age": 5,
+    "days_since_bind": 300,
+    "months_as_customer": 24,
+    "capital-gains": 0,
+    "capital-loss": 0,
+    "injury_share": 0.4,
+    "property_share": 0.6,
+    "umbrella_limit": 0,
+    "incident_hour_of_the_day": 14
+}

index.html

@@ -0,0 +1,417 @@
+<!DOCTYPE html>
+<html lang="en">
+
+<head>
+    <meta charset="UTF-8">
+    <meta name="viewport" content="width=device-width, initial-scale=1.0">
+    <title>Fraud Detection API - Client</title>
+    <style>
+        * {
+            margin: 0;
+            padding: 0;
+            box-sizing: border-box;
+        }
+
+        body {
+            font-family: 'Segoe UI', Tahoma, Geneva, Verdana, sans-serif;
+            background: linear-gradient(135deg, #667eea 0%, #764ba2 100%);
+            min-height: 100vh;
+            padding: 20px;
+            display: flex;
+            justify-content: center;
+            align-items: center;
+        }
+
+        .container {
+            background: white;
+            border-radius: 16px;
+            box-shadow: 0 20px 60px rgba(0, 0, 0, 0.3);
+            max-width: 900px;
+            width: 100%;
+            padding: 40px;
+        }
+
+        h1 {
+            color: #333;
+            margin-bottom: 10px;
+            font-size: 28px;
+        }
+
+        .subtitle {
+            color: #666;
+            margin-bottom: 30px;
+            font-size: 14px;
+        }
+
+        .config-section {
+            background: #f8f9fa;
+            padding: 20px;
+            border-radius: 8px;
+            margin-bottom: 30px;
+        }
+
+        .config-title {
+            font-weight: 600;
+            color: #495057;
+            margin-bottom: 15px;
+            font-size: 16px;
+        }
+
+        .config-grid {
+            display: grid;
+            grid-template-columns: repeat(auto-fit, minmax(200px, 1fr));
+            gap: 15px;
+        }
+
+        .form-group {
+            margin-bottom: 20px;
+        }
+
+        label {
+            display: block;
+            font-weight: 500;
+            margin-bottom: 5px;
+            color: #495057;
+            font-size: 14px;
+        }
+
+        input,
+        select {
+            width: 100%;
+            padding: 12px;
+            border: 2px solid #e9ecef;
+            border-radius: 6px;
+            font-size: 14px;
+            transition: border-color 0.3s;
+        }
+
+        input:focus,
+        select:focus {
+            outline: none;
+            border-color: #667eea;
+        }
+
+        .input-grid {
+            display: grid;
+            grid-template-columns: repeat(auto-fit, minmax(250px, 1fr));
+            gap: 15px;
+        }
+
+        .predict-btn {
+            background: linear-gradient(135deg, #667eea 0%, #764ba2 100%);
+            color: white;
+            border: none;
+            padding: 15px 40px;
+            border-radius: 8px;
+            font-size: 16px;
+            font-weight: 600;
+            cursor: pointer;
+            width: 100%;
+            margin-top: 20px;
+            transition: transform 0.2s, box-shadow 0.2s;
+        }
+
+        .predict-btn:hover {
+            transform: translateY(-2px);
+            box-shadow: 0 10px 25px rgba(102, 126, 234, 0.4);
+        }
+
+        .predict-btn:disabled {
+            opacity: 0.6;
+            cursor: not-allowed;
+            transform: none;
+        }
+
+        .result-section {
+            margin-top: 30px;
+            padding: 25px;
+            border-radius: 8px;
+            display: none;
+        }
+
+        .result-section.show {
+            display: block;
+        }
+
+        .result-section.fraud {
+            background: #fff5f5;
+            border: 2px solid #fc8181;
+        }
+
+        .result-section.legit {
+            background: #f0fff4;
+            border: 2px solid #68d391;
+        }
+
+        .result-title {
+            font-size: 20px;
+            font-weight: 600;
+            margin-bottom: 15px;
+        }
+
+        .result-grid {
+            display: grid;
+            grid-template-columns: repeat(auto-fit, minmax(200px, 1fr));
+            gap: 15px;
+        }
+
+        .result-item {
+            padding: 12px;
+            background: white;
+            border-radius: 6px;
+        }
+
+        .result-label {
+            font-size: 12px;
+            color: #718096;
+            text-transform: uppercase;
+            letter-spacing: 0.5px;
+            margin-bottom: 5px;
+        }
+
+        .result-value {
+            font-size: 18px;
+            font-weight: 600;
+            color: #2d3748;
+        }
+
+        .error-message {
+            background: #fff5f5;
+            border: 2px solid #fc8181;
+            color: #c53030;
+            padding: 15px;
+            border-radius: 8px;
+            margin-top: 20px;
+            display: none;
+        }
+
+        .error-message.show {
+            display: block;
+        }
+    </style>
+</head>
+
+<body>
+    <div class="container">
+        <h1>🛡️ Insurance Fraud Detection</h1>
+        <p class="subtitle">AI-powered fraud probability assessment for insurance claims</p>
+
+        <!-- Model Configuration -->
+        <div class="config-section">
+            <div class="config-title">Model Configuration</div>
+            <div class="config-grid">
+                <div class="form-group">
+                    <label for="model">Model Type</label>
+                    <select id="model">
+                        <option value="rf">Random Forest</option>
+                        <option value="et">Extra Trees</option>
+                        <option value="xgb" selected>XGBoost</option>
+                    </select>
+                </div>
+
+                <div class="form-group">
+                    <label for="scenario">Deployment Scenario</label>
+                    <select id="scenario">
+                        <option value="dashboard" selected>Dashboard (Calibrated)</option>
+                        <option value="auto_flagger">Auto-Flagger (Uncalibrated)</option>
+                    </select>
+                </div>
+            </div>
+        </div>
+
+        <!-- Claim Input Form -->
+        <form id="claimForm">
+            <div class="input-grid">
+                <div class="form-group">
+                    <label for="policy_annual_premium">Policy Annual Premium ($)</label>
+                    <input type="number" id="policy_annual_premium" step="0.01" value="1200" required>
+                </div>
+
+                <div class="form-group">
+                    <label for="total_claim_amount">Total Claim Amount ($)</label>
+                    <input type="number" id="total_claim_amount" step="0.01" value="15000" required>
+                </div>
+
+                <div class="form-group">
+                    <label for="vehicle_age">Vehicle Age (years)</label>
+                    <input type="number" id="vehicle_age" min="0" value="5" required>
+                </div>
+
+                <div class="form-group">
+                    <label for="days_since_bind">Days Since Policy Bind</label>
+                    <input type="number" id="days_since_bind" min="0" value="300" required>
+                </div>
+
+                <div class="form-group">
+                    <label for="months_as_customer">Months as Customer</label>
+                    <input type="number" id="months_as_customer" min="0" value="24" required>
+                </div>
+
+                <div class="form-group">
+                    <label for="injury_share">Injury Damage Share</label>
+                    <input type="number" id="injury_share" step="0.01" min="0" max="1" value="0.4" required>
+                </div>
+
+                <div class="form-group">
+                    <label for="property_share">Property Damage Share</label>
+                    <input type="number" id="property_share" step="0.01" min="0" max="1" value="0.6" required>
+                </div>
+
+                <div class="form-group">
+                    <label for="umbrella_limit">Umbrella Policy Limit</label>
+                    <input type="number" id="umbrella_limit" min="0" value="0" required>
+                </div>
+
+                <div class="form-group">
+                    <label for="incident_hour_of_the_day">Incident Hour (0-23)</label>
+                    <input type="number" id="incident_hour_of_the_day" min="0" max="23" value="14" required>
+                </div>
+
+                <div class="form-group">
+                    <label for="capital_gains">Capital Gains ($)</label>
+                    <input type="number" id="capital_gains" step="0.01" value="0">
+                </div>
+
+                <div class="form-group">
+                    <label for="capital_loss">Capital Loss ($)</label>
+                    <input type="number" id="capital_loss" step="0.01" value="0">
+                </div>
+            </div>
+
+            <button type="submit" class="predict-btn" id="predictBtn">
+                🔍 Analyze Claim
+            </button>
+        </form>
+
+        <!-- Result Display -->
+        <div id="resultSection" class="result-section">
+            <div class="result-title" id="resultTitle">Analysis Result</div>
+            <div class="result-grid">
+                <div class="result-item">
+                    <div class="result-label">Model Used</div>
+                    <div class="result-value" id="resultModel">-</div>
+                </div>
+                <div class="result-item">
+                    <div class="result-label">Fraud Probability</div>
+                    <div class="result-value" id="resultProbability">-</div>
+                </div>
+                <div class="result-item">
+                    <div class="result-label">Decision</div>
+                    <div class="result-value" id="resultDecision">-</div>
+                </div>
+                <div class="result-item">
+                    <div class="result-label">Scenario</div>
+                    <div class="result-value" id="resultScenario">-</div>
+                </div>
+            </div>
+        </div>
+
+        <!-- Error Display -->
+        <div id="errorMessage" class="error-message"></div>
+    </div>
+
+    <script>
+        const form = document.getElementById('claimForm');
+        const predictBtn = document.getElementById('predictBtn');
+        const resultSection = document.getElementById('resultSection');
+        const errorMessage = document.getElementById('errorMessage');
+
+        form.addEventListener('submit', async (e) => {
+            e.preventDefault();
+
+            // Hide previous results/errors
+            resultSection.classList.remove('show', 'fraud', 'legit');
+            errorMessage.classList.remove('show');
+
+            // Disable button
+            predictBtn.disabled = true;
+            predictBtn.textContent = '⏳ Analyzing...';
+
+            try {
+                // Gather form data
+                const formData = {
+                    policy_annual_premium: parseFloat(document.getElementById('policy_annual_premium').value),
+                    total_claim_amount: parseFloat(document.getElementById('total_claim_amount').value),
+                    vehicle_age: parseInt(document.getElementById('vehicle_age').value),
+                    days_since_bind: parseInt(document.getElementById('days_since_bind').value),
+                    months_as_customer: parseInt(document.getElementById('months_as_customer').value),
+                    'capital-gains': parseFloat(document.getElementById('capital_gains').value || 0),
+                    'capital-loss': parseFloat(document.getElementById('capital_loss').value || 0),
+                    injury_share: parseFloat(document.getElementById('injury_share').value),
+                    property_share: parseFloat(document.getElementById('property_share').value),
+                    umbrella_limit: parseInt(document.getElementById('umbrella_limit').value),
+                    incident_hour_of_the_day: parseInt(document.getElementById('incident_hour_of_the_day').value)
+                };
+
+                // Get model configuration
+                const model = document.getElementById('model').value;
+                const scenario = document.getElementById('scenario').value;
+
+                // Make API request
+                const response = await fetch(`/predict?model=${model}&scenario=${scenario}`, {
+                    method: 'POST',
+                    headers: {
+                        'Content-Type': 'application/json',
+                    },
+                    body: JSON.stringify(formData)
+                });
+
+                if (!response.ok) {
+                    const errorData = await response.json();
+                    throw new Error(errorData.detail || 'Prediction failed');
+                }
+
+                const result = await response.json();
+
+                // Display results
+                displayResults(result);
+
+            } catch (error) {
+                console.error('Error:', error);
+                errorMessage.textContent = `Error: ${error.message}`;
+                errorMessage.classList.add('show');
+            } finally {
+                // Re-enable button
+                predictBtn.disabled = false;
+                predictBtn.textContent = '🔍 Analyze Claim';
+            }
+        });
+
+        function displayResults(result) {
+            // Update result values
+            document.getElementById('resultModel').textContent =
+                `${result.model} ${result.calibrated ? '(Calibrated)' : '(Uncalibrated)'}`;
+
+            const probability = (result.probability * 100).toFixed(1);
+            document.getElementById('resultProbability').textContent = `${probability}%`;
+
+            // Determine decision text
+            let decision = '-';
+            if (result.threshold_flag) {
+                decision = result.threshold_flag === 'AUTO_FLAG' ?
+                    '🚨 FLAG FOR REVIEW' : '✅ AUTO APPROVE';
+            } else {
+                // For dashboard mode
+                if (result.probability >= 0.7) decision = '🔴 High Risk';
+                else if (result.probability >= 0.5) decision = '🟡 Medium Risk';
+                else decision = '🟢 Low Risk';
+            }
+            document.getElementById('resultDecision').textContent = decision;
+
+            document.getElementById('resultScenario').textContent =
+                result.scenario === 'auto_flagger' ? 'Auto-Flagger' : 'Dashboard';
+
+            // Style result section
+            resultSection.classList.add('show');
+            if (result.probability >= 0.5) {
+                resultSection.classList.add('fraud');
+                document.getElementById('resultTitle').textContent = '⚠️ High Fraud Risk Detected';
+            } else {
+                resultSection.classList.add('legit');
+                document.getElementById('resultTitle').textContent = '✓ Low Fraud Risk';
+            }
+        }
+    </script>
+</body>
+
+</html>

requirements.txt

@@ -0,0 +1,8 @@
+fastapi==0.104.1
+uvicorn[standard]==0.24.0
+pydantic==2.5.0
+joblib==1.3.2
+numpy==1.24.3
+pandas==2.0.3
+scikit-learn==1.3.2
+xgboost==2.0.3

test_api.sh

@@ -0,0 +1,56 @@
+#!/bin/bash
+
+# Fraud Detection API - Example curl Commands
+
+BASE_URL="http://localhost:7860"
+
+echo "========================================="
+echo "Fraud Detection API - Example Requests"
+echo "========================================="
+echo ""
+
+# Test 1: Dashboard Scenario (Calibrated) with XGBoost
+echo "1. Dashboard Scenario (Calibrated XGBoost):"
+echo "-----------------------------------------"
+curl -X POST "${BASE_URL}/predict?model=xgb&scenario=dashboard" \
+  -H "Content-Type: application/json" \
+  -d @example_claim.json
+echo -e "\n\n"
+
+# Test 2: Auto-Flagger Scenario (Uncalibrated) with RandomForest
+echo "2. Auto-Flagger Scenario (Uncalibrated RandomForest):"
+echo "-----------------------------------------------------"
+curl -X POST "${BASE_URL}/predict?model=rf&scenario=auto_flagger" \
+  -H "Content-Type: application/json" \
+  -d @example_claim.json
+echo -e "\n\n"
+
+# Test 3: High-Risk Claim Example
+echo "3. High-Risk Claim (Auto-Flagger with ExtraTrees):"
+echo "---------------------------------------------------"
+curl -X POST "${BASE_URL}/predict?model=et&scenario=auto_flagger" \
+  -H "Content-Type: application/json" \
+  -d '{
+    "policy_annual_premium": 500,
+    "total_claim_amount": 50000,
+    "vehicle_age": 1,
+    "days_since_bind": 10,
+    "months_as_customer": 2,
+    "capital-gains": 10000,
+    "capital-loss": 0,
+    "injury_share": 0.8,
+    "property_share": 0.2,
+    "umbrella_limit": 0,
+    "incident_hour_of_the_day": 3
+  }'
+echo -e "\n\n"
+
+# Test 4: Health Check
+echo "4. Health Check:"
+echo "----------------"
+curl -X GET "${BASE_URL}/health"
+echo -e "\n\n"
+
+echo "========================================="
+echo "All tests completed!"
+echo "========================================="