Optimization of Stochastic Inventory Models Using Machine Learning-Based Demand Prediction

Abstract

This study focuses on improving inventory management decisions under uncertain demand by integrating stochastic modeling with machine learning–based demand prediction. Traditional inventory systems often assume that demand follows a predictable pattern; however, in reality, it fluctuates due to seasonality, market dynamics, promotions, and supply disruptions. Machine learning provides a modern solution by analyzing large datasets and identifying nonlinear demand patterns that traditional models cannot capture. The proposed framework combines probabilistic demand forecasts from machine learning models with stochastic optimization techniques to minimize total inventory cost while maintaining desired service levels. The approach enables more accurate estimation of safety stock levels, reduces stockouts, and improves operational efficiency. Applications across industries such as aerospace, retail, logistics, and manufacturing demonstrate that machine learning–driven forecasting reduces holding costs, enhances responsiveness, and builds supply chain resilience. The study highlights how predictive analytics and optimization together create adaptive, data-driven inventory systems capable of performing effectively in uncertain and volatile business environments.