Your challenge in this sprint is improving the weekly sales forecasting models for the Christmas period. Your cross-validation strategy is ready, but before you can begin, you have to query the data from our systems and process them in a way that allows you to view the situation with clarity.
First, you have a meeting with Matthias, who’s worked on this problem before. During your meeting, you conclude that Christmas has a non-linear effect on sales. That’s why you decide to experiment with a multiplicative XGBoost in addition to your Regularised-Regression model. You make a grid with various features and parameters for both models and analyze the effects of both approaches. You notice your Regression is overfitting, which means XGBoost isn’t performing and the forecast isn’t high enough, so you increase the regularization and appoint the Christmas features to XGBoost alone.
Nice! You improved the precision of the Christmas forecast with an average of 2%. This will only yield results once the algorithm has been implemented, so you start thinking about how you want to implement this.