app.py

from fastapi import FastAPI, HTTPException
from pydantic import BaseModel, Field
import joblib
import json
import numpy as np
from datetime import datetime
from typing import Optional
import os

# Initialisation
app = FastAPI(
    title="API Détection de Fraude",
    description="API de détection de fraude dans les transactions bancaires",
    version="1.0.0"
)

# Chargement du modèle et des encodeurs
model = joblib.load('fraud_model.pkl')
le_category = joblib.load('le_category.pkl')
le_gender = joblib.load('le_gender.pkl')
le_state = joblib.load('le_state.pkl')

with open('mappings.json', 'r') as f:
    mappings = json.load(f)

# Modèle de données
class Transaction(BaseModel):
    amt: float = Field(..., description="Montant de la transaction", example=125.50)
    category: str = Field(..., description="Catégorie du marchand", example="personal_care")
    merchant: str = Field(..., description="Nom du marchand", example="fraud_Kirlin and Sons")
    trans_date_trans_time: str = Field(..., description="Date et heure", example="2020-06-21 12:14:25")
    gender: str = Field(..., description="Genre (M/F)", example="M")
    state: str = Field(..., description="État (code à 2 lettres)", example="SC")
    lat: float = Field(..., description="Latitude du client", example=33.9659)
    long: float = Field(..., description="Longitude du client", example=-80.9355)
    city_pop: int = Field(..., description="Population de la ville", example=333497)
    dob: str = Field(..., description="Date de naissance", example="1968-03-19")
    merch_lat: float = Field(..., description="Latitude du marchand", example=33.986391)
    merch_long: float = Field(..., description="Longitude du marchand", example=-81.200714)
    cc_num: Optional[str] = Field(None, description="Numéro de carte (optionnel)")
    avg_amt: Optional[float] = Field(50.0, description="Montant moyen historique")
    std_amt: Optional[float] = Field(30.0, description="Écart-type historique")
    nb_trans: Optional[int] = Field(10, description="Nombre de transactions historiques")

class PredictionResponse(BaseModel):
    is_fraud: bool
    fraud_probability: float
    risk_level: str
    details: dict

# Routes
@app.get("/")
def read_root():
    return {
        "message": "API de Détection de Fraude",
        "version": "1.0.0",
        "endpoints": {
            "/predict": "POST - Prédire une transaction",
            "/health": "GET - Statut de l'API",
            "/categories": "GET - Liste des catégories",
            "/docs": "GET - Documentation interactive"
        }
    }

@app.get("/health")
def health_check():
    return {
        "status": "healthy",
        "model_loaded": model is not None,
        "timestamp": datetime.now().isoformat()
    }

@app.get("/categories")
def get_categories():
    return {
        "categories": list(mappings['categories'].keys()),
        "states": list(mappings['states'].keys()),
        "genders": list(mappings['genders'].keys())
    }

@app.post("/predict", response_model=PredictionResponse)
def predict_fraud(transaction: Transaction):
    try:
        # Parsing des dates
        trans_dt = datetime.strptime(transaction.trans_date_trans_time, "%Y-%m-%d %H:%M:%S")
        dob_dt = datetime.strptime(transaction.dob, "%Y-%m-%d")
        
        # Features temporelles
        hour = trans_dt.hour
        day_of_week = trans_dt.weekday()
        day = trans_dt.day
        month = trans_dt.month
        age = (trans_dt - dob_dt).days // 365
        
        # Distance
        distance = np.sqrt(
            (transaction.lat - transaction.merch_lat)**2 + 
            (transaction.long - transaction.merch_long)**2
        ) * 111
        
        # Encodage
        if transaction.category not in mappings['categories']:
            raise HTTPException(400, f"Catégorie inconnue: {transaction.category}")
        if transaction.gender not in mappings['genders']:
            raise HTTPException(400, f"Genre inconnu: {transaction.gender}")
        if transaction.state not in mappings['states']:
            raise HTTPException(400, f"État inconnu: {transaction.state}")
        
        category_encoded = mappings['categories'][transaction.category]
        gender_encoded = mappings['genders'][transaction.gender]
        state_encoded = mappings['states'][transaction.state]
        
        # Construction du vecteur de features
        features = np.array([[
            transaction.amt,
            hour,
            day_of_week,
            day,
            month,
            age,
            category_encoded,
            gender_encoded,
            state_encoded,
            transaction.lat,
            transaction.long,
            transaction.city_pop,
            distance,
            transaction.avg_amt,
            transaction.std_amt,
            transaction.nb_trans
        ]])
        
        # Prédiction
        fraud_proba = model.predict_proba(features)[0][1]
        is_fraud = fraud_proba > 0.5
        
        # Niveau de risque
        if fraud_proba < 0.3:
            risk_level = "Faible"
        elif fraud_proba < 0.7:
            risk_level = "Moyen"
        else:
            risk_level = "Élevé"
        
        return PredictionResponse(
            is_fraud=bool(is_fraud),
            fraud_probability=float(fraud_proba),
            risk_level=risk_level,
            details={
                "montant": transaction.amt,
                "categorie": transaction.category,
                "heure": hour,
                "age_client": age,
                "distance_km": round(distance, 2)
            }
        )
        
    except ValueError as e:
        raise HTTPException(status_code=400, detail=f"Erreur de format: {str(e)}")
    except Exception as e:
        raise HTTPException(status_code=500, detail=f"Erreur interne: {str(e)}")

if __name__ == "__main__":
    import uvicorn
    uvicorn.run(app, host="0.0.0.0", port=7860)