In this repo, you will find my XGBoost model and a documentation of the feature extraction I used for sales forecasting. I used the Walmart Sales Prediction Kaggle Dataset, which contains sales data for different stores and departments from 05-Feb-2010 to 26-Oct-2012.
My model achieved a mean absolute percentage error of 2.36% over 7 months of hold out data.
The XGBoost model is then compared against forecasts using Holt Winter's method.
Read the full Jupyter Notebook at https://kevinchtsang.github.io/walmart-sales/