saniamirza/supply-chain-ml-models

Supply Chain ML Models

Machine Learning models for supply chain risk analysis.

Models

This repository contains three trained ML components:

1. Delivery Delay Prediction

Random Forest Classifier used to estimate the probability that an order will be delivered late.

Output:

• delivery_risk

2. Anomaly Detection

Isolation Forest used to identify unusual supply-chain orders.

Outputs:

• anomaly_prediction
• anomaly_score
• anomaly_risk

3. Demand Forecasting

Random Forest Regressor used to forecast monthly demand using historical demand features.

Output:

• predicted demand

Supply Chain Risk

Delivery Risk and Anomaly Risk are combined to generate an overall Supply Chain Risk Score.

The final risk score is calculated using:

Supply Chain Risk = 60% Delivery Risk + 40% Anomaly Risk

Risk categories:

• Low: score < 25
• Medium: score 25–60
• High: score > 60

Usage

Install dependencies:

pip install -r requirements.txt

Supply Chain Risk Prediction

Import the prediction function:

from predict import predict_supply_chain_risk

The predict_supply_chain_risk() function requires the following input fields:

• product_id
• customer_id
• customer_segment
• sales
• quantity
• shipping_mode
• market
• lead_time
• avg_order_value_30d
• num_orders_30d
• is_high_value
• is_bulk_order
• day_of_week
• month
• quarter
• year
• department
• class
• profit
• order_processing_days
• avg_lead_time_by_mode
• avg_shipping_cost
• avg_defect_rate
• max_defect_rate
• profit_margin

Provide the required order data:

order_data = {
    "product_id": 365,
    "customer_id": 2,
    "customer_segment": "Consumer",
    "sales": 119.98,
    "quantity": 2,
    "shipping_mode": "Standard Class",
    "market": "LATAM",
    "lead_time": 10,
    "avg_order_value_30d": 119.98,
    "num_orders_30d": 1,
    "is_high_value": 0,
    "is_bulk_order": 0,
    "day_of_week": 3,
    "month": 1,
    "quarter": 1,
    "year": 2017,
    "department": "Technology",
    "class": "Regular Air",

    "profit": 20.0,
    "order_processing_days": 2,
    "avg_lead_time_by_mode": 10.0,
    "avg_shipping_cost": 5.0,
    "avg_defect_rate": 1.0,
    "max_defect_rate": 2.0,
    "profit_margin": 0.17
}

result = predict_supply_chain_risk(order_data)

print(result)

Example output:

{
    "delivery_risk": 22.0,
    "anomaly_prediction": "Normal",
    "anomaly_score": 0.059487,
    "anomaly_risk": 40.06,
    "supply_chain_risk": 29.23,
    "risk_category": "Medium"
}

Demand Forecasting

Import the demand forecasting function:

from predict import predict_demand

The predict_demand() function requires:

• lag_1
• lag_2
• lag_3
• month

Provide the previous three months of demand:

prediction = predict_demand(
    lag_1=4675,
    lag_2=4146,
    lag_3=4823,
    month=10
)

print(prediction)

Example output:

4701.02

Project Structure

supply-chain-ml-models/
│
├── models/
│   ├── delivery_delay_model.pkl
│   ├── delivery_preprocessor.pkl
│   ├── demand_forecasting_model.pkl
│   ├── demand_forecast_features.pkl
│   ├── anomaly_detection_model.pkl
│   ├── anomaly_scaler.pkl
│   └── anomaly_risk_scaler.pkl
│
├── predict.py
├── requirements.txt
└── README.md

Backend Integration

The predict.py module is designed to be imported by a backend API.

The backend can call:

from predict import predict_supply_chain_risk
from predict import predict_demand

For supply chain risk:

result = predict_supply_chain_risk(order_data)

For demand forecasting:

prediction = predict_demand(
    lag_1,
    lag_2,
    lag_3,
    month
)

The returned prediction results can be converted to JSON by the backend API and sent to the frontend.

Model Testing

The trained models were independently tested after saving.

Tested components:

• Delivery delay prediction
• Anomaly detection
• Demand forecasting
• Supply chain risk calculation

The models were also tested after downloading them from the Hugging Face repository.

The downloaded models successfully loaded and generated predictions through the standalone inference pipeline.

Hugging Face Repository

The trained models and inference code are hosted in this repository:

saniamirza/supply-chain-ml-models

The repository contains all required model files, preprocessors, scalers, inference code, dependencies, and documentation for backend integration.