Optimizing Supply Chain Performance: A Hybrid Approach that Integrates Artificial Intelligence and Mathematical Modeling with Stochastic Models to Reduce Costs and Improve Efficiency in Manufacturing

Abstract

Supply chain management plays a role in guaranteeing cost-efficiency and the smooth running of production and logistics. Existing methods, such as deterministic optimization models and traditional machine learning approaches, often face limitations in handling uncertainties, temporal dependencies, and high-dimensional structured data, which restrict their ability to achieve robust forecasting and optimization. To address these concerns, an integrated framework using the TabNet and Stochastic Mixed-Integer Linear Programming (SMILP) is proposed with an aim to cost reduction and increasing efficiency in supply chain systems. The methodology proposed herein initiates with the preprocessing of data, with missing values in the dataset imputation process using K-Nearest Neighbor (KNN) and mode imputation methods, including Min–Max scale normalization. Next, feature engineering is conducted, where rolling window statistics are applied to provide seasonal and temporal demand trend information. The TabNet model is used to learn feature importance and provide demand forecasting, which is then routed into SMILP to optimization production, inventory, and logistics in anticipation of uncertainty. The hybrid orientation encapsulates the predictive power of TabNet, as well as the power to make decisions in the presence of stochastic optimization, to provide resiliency and adaptability. After evaluating the effectiveness with Mean Absolute Error (MAE), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE), we conclude that the cost-optimizing power of the model affects management in the supply chain.