Predict and Optimize Inventory Management Decisions for Beauty Retail

This thesis develops an adaptive inventory management framework that integrates machine learning forecasting with stochastic optimization to address the challenges of uncertain, dynamic retail demand. Using data from a major Australian beauty retailer, it examines three settings: continuous review systems with daily forecast updates, periodic review systems with fixed replenishment schedules and operational constraints, and forecasting for new products without sales history by identifying comparable items and adjusting for seasonality. By generating demand scenarios and updating decisions in response to new information, the proposed methods deliver more accurate forecasts, improved service levels, and lower inventory costs than classical models and expert judgment.