M5 Demand Forecasting (Walmart)

This project demonstrates an end-to-end retail demand forecasting pipeline using the M5 dataset.

Problem

Forecast daily unit sales for the next 28 days for a Walmart item-store series.

Approach

• Exploratory Data Analysis to understand intermittent demand
• Strong baseline forecasting (naive, 7-day mean, 28-day mean)
• Leakage-safe lag and rolling feature engineering
• LightGBM regression with time-based validation
• Hybrid approach blending ML predictions with rolling mean baselines
• Recursive 28-day demand forecasting

Example Forecast

Key Insight

For intermittent retail demand, predicting expected demand levels is more reliable than chasing rare spikes. Blending statistical baselines with ML improves forecast stability and realism.

Tools

Python, Pandas, NumPy, LightGBM, Matplotlib