Publication Announcement from Pedro Cesar Lopes Gerum

Improving High-frequency Demand Forecasts for Omnichannel Grocery Retail 

In the fast-paced world of online grocery retail, predicting demand for specific delivery windows is notoriously challenging. Traditional forecasting models often overemphasize recent data, overlooking important long-term patterns such as holidays and seasonal trends. In a new study published in the International Journal of Forecasting—the premier journal in the field—Pedro Cesar Lopes Gerum and co-author Moonwon Chung (former CSU professor) introduce an innovative “decoupling” approach to address this issue.

The paper, titled "Improving High-Frequency Demand Forecasts for Omnichannel Grocery Retail," splits the forecasting problem into two distinct components. It utilizes Mixture Density Networks (MDN) to capture long-term, multimodal trends and integrates them with N-HiTS to model short-term volatility. When implemented with a major European grocery retailer, this method achieved a 37% reduction in forecasting error, directly improving inventory management and driver scheduling efficiency.

Click here to read the paper on Science Direct 

Watch as Pedro explains his research and its real-world impact.