Danang08/data_mining

1from fastapi import FastAPI2from fastapi.middleware.cors import CORSMiddleware3from pydantic import BaseModel, Field4from catboost import CatBoostClassifier5import pandas as pd6import joblib7import os8 9MODEL_PATH = os.path.join("model_output", "catboost_return_model.cbm")10META_PATH = os.path.join("model_output", "model_metadata.pkl")11 12model = CatBoostClassifier()13model.load_model(MODEL_PATH)14metadata = joblib.load(META_PATH)15 16app = FastAPI(17    title="E-Commerce Return Risk Prediction API",18    description="Backend prediksi retur e-commerce menggunakan CatBoost dari notebook Colab UTS Danang.",19    version="1.0.0"20)21 22# Untuk tugas kuliah dibuat terbuka. Kalau production, batasi origin domain Vercel kamu.23app.add_middleware(24    CORSMiddleware,25    allow_origins=["*"],26    allow_credentials=True,27    allow_methods=["*"],28    allow_headers=["*"],29)30 31class TransactionInput(BaseModel):32    product_category: str = Field(..., example="Electronics")33    product_price: float = Field(..., example=250.0)34    quantity: int = Field(..., example=2)35    order_date: str = Field(..., example="2025-05-15")36    region: str = Field(..., example="Asia")37    payment_method: str = Field(..., example="CreditCard")38    delivery_days: int = Field(..., example=5)39    customer_rating: float = Field(..., example=4.2)40    discount_percent: float = Field(..., example=10.0)41    revenue: float = Field(..., example=450.0)42 43def create_features(input_dict: dict) -> pd.DataFrame:44    data = pd.DataFrame([input_dict])45 46    data["order_date"] = pd.to_datetime(data["order_date"], errors="coerce")47    data["order_year"] = data["order_date"].dt.year48    data["order_month"] = data["order_date"].dt.month49    data["order_day"] = data["order_date"].dt.day50    data["order_dayofweek"] = data["order_date"].dt.dayofweek51    data["is_weekend"] = data["order_dayofweek"].isin([5, 6]).astype(int)52    data["order_quarter"] = data["order_date"].dt.quarter53 54    data["price_after_discount"] = data["product_price"] * (1 - data["discount_percent"] / 100)55    data["estimated_total_price"] = data["price_after_discount"] * data["quantity"]56 57    data = data.drop(columns=["order_date"], errors="ignore")58 59    # Pastikan urutan kolom sama persis dengan saat training di Colab60    data = data[metadata["features"]]61    return data62 63@app.get("/")64def root():65    return {66        "message": "API Prediksi Retur E-Commerce aktif.",67        "model": "CatBoostClassifier",68        "threshold": metadata["best_threshold"],69        "features": metadata["features"]70    }71 72@app.get("/health")73def health():74    return {"status": "ok"}75 76@app.get("/metadata")77def get_metadata():78    return metadata79 80@app.post("/predict")81def predict(payload: TransactionInput):82    input_data = create_features(payload.model_dump())83 84    probability = float(model.predict_proba(input_data)[:, 1][0])85    threshold = float(metadata["best_threshold"])86    prediction = int(probability >= threshold)87 88    if probability >= 0.70:89        risk_level = "Tinggi"90    elif probability >= 0.40:91        risk_level = "Sedang"92    else:93        risk_level = "Rendah"94 95    return {96        "probability_return": round(probability, 4),97        "probability_percent": round(probability * 100, 2),98        "threshold": round(threshold, 2),99        "prediction": prediction,100        "label": "Risiko Retur Tinggi" if prediction == 1 else "Risiko Retur Rendah",101        "risk_level": risk_level102    }103