prediction.py

import joblib
import pandas as pd
import numpy as np
from scipy.stats import poisson

model = joblib.load("inventory_model.pkl")  # Load trained model

INDUSTRIAL_PARAMS = {
    "service_levels": {"A": 2.33, "B": 1.65, "C": 1.28},
    "risk_thresholds": {"Low": 15, "Medium": 40, "High": 100}
}

def predict_stockout(order_quantity: float, lead_time: float, safety_stock: float, product_class: str) -> str:
    if not (0 <= order_quantity <= 10000 and 0 <= lead_time <= 90 and 0 <= safety_stock <= 10000):
        return "Error: Inputs out of range (Demand: 0-10,000, Lead Time: 0-90 days, Safety Stock: 0-10,000)."
    if any(x < 0 for x in [order_quantity, lead_time, safety_stock]):
        return "Error: Negative values not allowed."
    if product_class not in INDUSTRIAL_PARAMS["service_levels"]:
        return "Error: Invalid product class. Use A, B, or C."

    daily_demand = order_quantity / 30
    expected_demand = daily_demand * lead_time
    poisson_risk = max(0, min(1, 1 - poisson.cdf(safety_stock, max(1, expected_demand))))
    
    safety_stock_impact = safety_stock / (order_quantity + 1e-6)
    lead_time_impact = lead_time / 90
    order_quantity_impact = order_quantity / 10000
    
    ml_input = pd.DataFrame({"safety_stock_impact": [safety_stock_impact], "lead_time_impact": [lead_time_impact], "order_quantity_impact": [order_quantity_impact]})
    ml_risk = max(0, min(1, model.predict(ml_input)[0]))
    
    weight_poisson = min(1, max(0, 1 - (order_quantity / 1000)))
    weight_ml = 1 - weight_poisson
    blended_risk = (weight_poisson * poisson_risk + weight_ml * ml_risk)
    risk_percentage = np.clip(blended_risk * 100, 0, 100)
    
    thresholds = INDUSTRIAL_PARAMS["risk_thresholds"]
    if risk_percentage < thresholds["Low"]:
        return f"<div class='risk-low'>Low Risk (Class {product_class})<br>Risk: {risk_percentage:.1f}%<br>No action needed</div>"
    elif risk_percentage < thresholds["Medium"]:
        return f"<div class='risk-medium'>Medium Risk (Class {product_class})<br>Risk: {risk_percentage:.1f}%<br>Monitor closely</div>"
    else:
        return f"<div class='risk-high'>High Risk (Class {product_class})<br>Risk: {risk_percentage:.1f}%<br>Replenish inventory</div>"  # Return risk prediction HTML