Inventory Demand Forecasting Using XGBoost and LightGBM Algorithms: A Case Study of Grupo Bimbo

Abstract

In the modern global economy, the complexity of supply chains is on the rise, making demand forecasting crucial yet challenging due to unpredictable elements such as seasonality, market dynamics, and economic variables. Conventional forecasting approaches are frequently unable to detect subtle yet critical trends. Conversely, machine learning techniques offer more accurate demand predictions by analyzing patterns from diverse data sources. In this research, two algorithms including XGBoost and LightGBM are applied to forecast inventory demand using the Grupo Bimbo Inventory Demand dataset. The models are then compared and evaluated using three metrics: RMLSE, MAE, and R2. The study reveals that the LightGBM model had a lower RMLSE (0.09954), while XGBoost model performed better in MAE (50.57126) and R2 (0.9821). Furthermore, the findings highlight that attributes such as sales unit this week and prior week's inventory demand significantly impacted forecasting accuracy.