<![CDATA[Sales forecasting in retail remains a challenge, as changing customer behavior and seasonal patterns undermine predictive accuracy. The Ensemble Stacking Framework for Machine Learning & Statistical Models is being suggested to improve forecasting of dynamic demands. For example, during promotional events, managers adjust their plans to account for changes in weather or unexpected changes in consumer preferences. They are unable to predict these changes. An Ensemble Stacking Framework was created using a Meta-Learner (RidgeCV) trained on historical retail sales data and several base learners (Random Forest, Extra Trees, XGBoost, and Multiple Linear Regression). Tree-Based Models help to create non-linear relationships, while Linear Regression helps us understand trend information. The Meta-Learner learns how to weigh each model based on the error each model makes over time. We assessed the quality of our model’s performance through RMSE and MAE. Special attention was given to assessing how well our models performed during periods of high volatility (i.e. holidays and post-promotional dips). Our model performed better than all other Base Models at all times. Also, stacked predictions showed less variation week-to-week and were able to return to normal quicker than other models after sudden shifts in demand. In practice, it allows for more confident safety-stock decisions and timely replenishment adjustments. Retailers can run a number of models in parallel and continuously rebalance them with new data as it comes in. Future work may include online updates of meta-learners and external covariates such as local events or competitor pricing.]]>
Copyrights © 2026
