Project 01: Business Sales Prediction

Objective

Build a machine learning system that predicts future sales from historical business data and explains model performance clearly.

Learning Outcomes

• Clean structured datasets.
• Create features from time, product, customer, price, and promotion fields.
• Train and compare classical ML models.
• Evaluate regression quality.
• Serve predictions through an API.
• Document limitations and next improvements.

Suggested Dataset Fields

• Date.
• Product ID.
• Customer segment.
• Sales quantity.
• Sales amount.
• Price.
• Discount.
• Promotion flag.
• Channel.
• Seasonality indicators.

Architecture

Raw Data -> Cleaning -> Feature Engineering -> Train/Validation/Test -> Model Training -> Evaluation -> API -> Dashboard or Report

Tasks

• Create dataset description.
• Perform exploratory data analysis.
• Build a baseline model.
• Train linear regression, random forest, and gradient boosting.
• Compare MAE, RMSE, and MAPE.
• Create feature importance analysis.
• Build a prediction API.
• Add Dockerfile.
• Write README and evaluation report.

Deliverables

• README.md
• docs/architecture.md
• docs/evaluation.md
• notebooks/exploration.ipynb
• src/ training and inference code
• tests/ basic tests
• Dockerfile

Tips

• Start with a naive baseline such as last-period sales.
• Keep temporal validation separate from random validation.
• Explain business meaning, not only metric values.
• Document when the model should not be trusted.