Machine Learning in Prescribing Optimal Target Inventory Levels for FMCG

For our project sponsor, a leading global Fast Moving Consumer Goods (FMCG) company, the goal goes beyond simply meeting complex demands. The focus is on optimizing inventory management strategies to balance the crucial trade-off between avoiding stockouts and minimizing the costs associated with excess inventory. Our project explores machine learning's potential to enhance the company's strategy for setting target inventory levels. We aim to use machine learning models to determine how much optimization is possible, ensuring that product service levels are sufficient to meet demand. Utilizing company data and external variables as features, we developed and evaluated various machine learning techniques by tuning different hyperparameter sets to identify the most accurate binary classification model. This model is designed to accurately predict stockout events for each stock keeping unit (SKU) across the company's warehouses. We selected the most accurate model based on a classification report that included key metrics such as precision, recall, f1-score, and accuracy. Using this model, we conducted sensitivity analysis to test different scenarios of adjusting current target inventory levels and assess how these changes could affect predicted stockout events. The findings will offer the company recommendations on setting inventory levels for specific SKUs at warehouse sites, ensuring that target service levels are maintained.