Optimizing Multi-Tier Supply Chain Ordering with a Hybrid Liquid Neural Network and Extreme Gradient Boosting Model

Abstract Supply chain management (SCM) encounters major obstacles, such as demand variability, inventory mismatches, and increased upstream order fluctuations caused by the bullwhip effect. Conventional approaches, like simple moving averages, find it challenging to cope with ever-changing market conditions. The Vending Machine Test is a key benchmark for LLMs, simulating real-world vending machine sales prediction scenarios, but Grok-4 and many other AI models still struggle with the complex continuous time series data in SCM, making it hard to accurately predict vending machine sales that fluctuate hourly, daily or seasonally. However, new machine learning (ML) methods, including LSTM, reinforcement learning, and XGBoost, present possible solutions but are hindered by computational complexity, training inefficiencies, or limitations in time-series modeling. Liquid Neural Networks (LNN), drawing inspiration from dynamic biological systems, offer a promising alternative due to their adaptability, low computational demands, and resilience to noise, making them ideal for real-time decision-making and edge computing. Although they have been successful in areas like autonomous vehicles and medical monitoring, their potential in optimizing supply chains is still largely untapped. This study introduces a hybrid LNN + XGBoost model to refine ordering strategies in multi-tier supply chains. By utilizing LNN’s dynamic feature extraction and XGBoost’s global optimization strengths, the model seeks to reduce the bullwhip effect and boost overall profitability. The research explores how the hybrid framework’s local and global synergies meet the dual needs of adaptability and efficiency in SCM. The proposed method addresses a crucial gap in current methodologies, providing an innovative solution for dynamic and efficient supply chain management.