A comparative study of multi-algorithm optimization for inventory analytics in supply chains

Effective management of inventory is essential for achieving high service levels, minimizing costs, and maintaining the overall resilience of retail supply chains—particularly in complex, real-world environments. Conventional strategies often prove inadequate because they rely on rigid assumptions or single-technique models that fail to accommodate practical challenges such as fluctuating demand, unpredictable lead times, and disruptions in supply. To bridge this gap, our research undertakes a comprehensive comparison of multiple approaches—including Reinforcement Learning (RL), Genetic Algorithms (GA), Deep Learning (DL), Machine Learning (ML), and heuristic techniques—evaluated within a consistent and realistic testing framework based on the Walmart M5 dataset. This dataset offers a robust benchmark, containing multi-store, multi-item sales data that captures seasonal trends, event-driven demand variations, and price sensitivity. We introduce and evaluate an innovative hybrid methodology that combines a Genetic Algorithm with a Deep Q-Network (GA–DQN). The GA component conducts a broad, global search to optimize static inventory parameters such as reorder points and safety stock, while the DQN module learns adaptive, state-aware ordering strategies that can respond to dynamic, uncertain conditions. Our results show that this hybrid GA–DQN model achieves a significant improvement over a standalone DQN baseline—raising the service level from 61% to 94% and simultaneously lowering overall inventory costs. The framework we propose is modular and includes three key components: demand forecasting using Long Short-Term Memory (LSTM) networks to capture temporal sales patterns; GA-based optimization to fine-tune static policy parameters; and RL-driven adaptive control to support responsive, real-time ordering decisions. This integrated approach delivers a scalable, data-driven solution well-suited to the demands of modern retail supply chains, effectively addressing issues such as supplier unreliability, demand uncertainty, and the management of perishable goods. • Demonstrate a hybrid framework combining deep learning, evolutionary search, and reinforcement learning for inventory optimization. • Benchmark diverse optimization methods using real-world retail data for comprehensive supply chain analytics. • Achieve superior inventory performance with a hybrid approach, improving service levels and reducing costs. • Integrate forecasting, optimization, and adaptive control into a modular system for scalable supply chain solutions. • Evaluate framework robustness against demand fluctuations, supplier variability, and supply chain disruptions.