AI-Driven Inventory and Supply Chain Optimization: A Comparative Review of Tree-Based Machine Learning and Sequence Deep Learning Models

Abstract

Artificial intelligence (AI) is essential in supply chain and inventory management due to the increasing complexity, volatility, and unpredictability of these processes.  Demand variations, seasonality, and nonlinear dependencies are frequently missed by conventional statistics and optimization techniques.  Models for deep learning (DL) and machine learning (ML) have lately demonstrated significant promise for raising predicting accuracy, lowering costs, and enhancing robustness.  Tree-based machine learning techniques like Random Forest, XGBoost, LightGBM, and CatBoost are prized for their effectiveness and interpretability when working with structured data. While they require more computing power, sequential deep learning models such as Temporal Convolution Networks (TCN) and Long Short-Term Memory (LSTM) are superior at simulating temporal dependencies and generating precise long-term predictions. This study uses the PRISMA methodology to comprehensively analyze 42 peer-reviewed articles that were published between 2019 to August 2025.  With the help of metrics like RMSE, MAE, and MAPE, the comparison assesses forecasting accuracy, scalability, interpretability, and processing overhead.  The results show the trade-offs between the two methods and point out areas that need more research, such as the dearth of integrated comparative studies and the scant attention given to hybrid models.  Amazon, Walmart, and Alibaba real-world examples demonstrate the usefulness of these techniques. These results highlight the potential of explainable AI and hybrid architectures to bridge the gap between theory and practice.  For scholars and industry professionals looking for scalable, data-driven, and robust supply chain solutions, this evaluation offers insightful information.