Data engineering
Sales, pricing, and calendar pipelines.
← Selected work
Machine learning
Retail demand forecasting
28-day forecasts across 30,490 product-store series.
The work
Temporal features
Lag and rolling features without future-data leakage.
Model evaluation
LightGBM and classical forecasting baselines.
Measured performance
23.8% lower mean RMSSE than the weekly seasonal-naive baseline.
LightGBM achieved 0.7595 mean RMSSE in chronological evaluation on the M5 dataset. Results describe forecasting accuracy in this project.