Abstract
ABSTRACT The retail sector faces growing challenges, particularly aligning with the European Union’s sustainable policies to minimize waste. This paper proposes a framework to address two pivotal goals: i) Introducing cutting-edge machine learning models to forecast demand within the context of a post-COVID environment. ii) Evaluating the benefits of integrating these predictive into operational strategies by measuring the reduction in overstock levels compared to traditional business practices. The hybrid Prophet-XGBoost model consistently outperformed classical and other hybridization models in terms of accuracy (lowest MAPE and WAPE), when predicting demand. This study uses data from 2019 to 2023 but excludes 2020 and 2021 due to the disruptions caused by COVID-19. Our findings reveal that relying solely on recent data from 2022 to 2023 results in lower model accuracy compared to historical imputation methods. Notably, substituting 2019 values for 2021 outperforms interpolating with data from 2022. Beyond its methodological advancements, this research introduces a novel approach to quantifying overstock reduction, contributing to both academic literature and retail practice. In this case, we observed a significant overstock issue with non-food products, likely tied to agreements between retailers and suppliers. As these products are non-perishable, retailers appear to have been less cautious in managing stock levels.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call