A machine learning system that predicts optimal markup percentages for product pricing based on order characteristics and customer attributes.
This project demonstrates how to build a pricing optimization model using historical order data. The model predicts the appropriate markup percentage for each order based on:
- Order features: delivery distance, quantity, unit cost, packaging type
- Customer features: industry, credit limit, payment terms
The dataset contains ~12,800 orders and ~735 customers with the following key columns:
Order Data (order_df.csv):
| Column | Description |
|---|---|
delivery_distance |
Distance in miles |
order_quantity_lbs |
Order quantity in pounds |
average_unit_cost_lbs |
Unit cost in USD/lb |
packaging |
Packaging type (BULK, -MDRM) |
markup_percentage |
Target variable |
customer_id |
Link to customer data |
Customer Data (customer_df.csv):
| Column | Description |
|---|---|
industry_name |
Customer industry (Pharma, Food, etc.) |
credit_limit |
Customer credit limit |
term_code |
Payment terms |
| Model | Test R² | Notes |
|---|---|---|
| Linear Regression | 0.63 | Baseline model |
| Decision Tree | 0.75 | Depth limited to 5 |
| Random Forest | 0.87 | 100 estimators, depth 10 |
| XGBoost | 0.88 | Best performance with customer features |
# Install dependencies
pip install -r requirements.txt
# Run the notebook
jupyter notebook pricing_optimization.ipynb├── notebooks/
│ └── pricing_optimization.ipynb # Main analysis notebook
├── data/
│ ├── customer_df.csv # Customer attributes
│ ├── order_df.csv # Historical orders
│ └── inference/
│ ├── inference_df.csv # New orders for prediction
│ └── inference_predicted.csv # Predictions output
├── requirements.txt
└── README.md
- Customer industry matters: Chemical Processing and Pharma customers typically receive different markup rates
- Order quantity impacts pricing: Larger bulk orders tend to have different markup strategies
- Delivery distance affects costs: Longer distances increase delivery costs which impacts optimal markup
The trained model can be used to predict markup percentages for new orders:
# Load the model and predict
import joblib
model = joblib.load('xgboost_model.pkl')
predicted_markup = model.predict(new_order_features)
# Calculate selling price
selling_price = unit_cost * (1 + predicted_markup)MIT License