Hybrid Machine Learning Optimization for Production Scheduling and Resource Al location in Smart Manufacturing Systems: A Review

Efficient resource allocation and production scheduling directly dictate manufacturing productivity, cost efficiency, and energy utilization. However, intensifying sustainability mandates, operational uncertainty, and demand volatility have exposed the limits of traditional, deterministic optimization methods. Consequently, contemporary research has pivoted toward integrating mathematical programming with Machine Learning (ML) to enhance decision-making under uncertainty. This review synthesizes literature from 2020 to 2026 across standalone classical optimization, ML approaches, and hybrid ML-optimization frameworks, critically evaluating their performance outcomes, archi tectural trends, and robustness across diverse manufacturing systems. The findings indicate that while mathematical models ensure solution transparency and strict constraint enforcement, ML components maximize adaptability to dynamic conditions and demand forecasting precision. While hybrid frameworks generally yield superior makespan reduction and schedule robustness, their efficacy is bounded by distinct computational latency and data-density trade-offs. Additionally, multi-tiered sensitivity and scenario analyses are established as vital diagnostics for validat ing model stability against stochastic shopfloor noise and macro-systemic shocks. Despite these analytical advances, significant deployment barriers persist regarding computational complexity, interpretability, real-time scalability, and data quality. To bridge these gaps, this review maps emerging frontiers in Industry 5.0 human-centric optimiza tion, reinforcement learning, and digital twins, providing a structured foundation for deploying resilient, scalable, and sustainable production planning systems.