Grocery Sales Forecasting
Feature engineering and classical forecasting methods for the Corporación Favorita dataset.
Overview
A learning-oriented reproduction of classical approaches to the Corporación Favorita grocery sales forecasting task. The work follows the complete process from exploratory analysis and feature construction to model fitting and evaluation.
The implementation combines historical sales, planning variables, and auxiliary information in an XGBoost forecasting pipeline. It studies a 16-day prediction horizon across products and stores.
Attribution and materials
The approach draws on prior competition solutions, which are linked and credited in the repository README. Read the project report.