Sustainable Optimization in Supply Chain Management Using Machine Learning
Junlin Zhu, Ying Wu, Zihao Liu, Cecelia Costa
Abstract
The complexity of modern supply chains requires robust solutions to improve efficiency, resilience and sustainability. This study proposes a systematic data-driven framework that integrates predictive analytics, anomaly detection, and multi-objective optimization. The enhanced LSTM model features an attention mechanism that reduces RMSE by up to 18.8%, significantly improves demand forecast accuracy and reduces the risk of out-of-stock and over-stocking. Anomaly detection showed a true positive rate exceeding 95%, reducing response times to disruptions by 28.5%. In addition, resource optimization resulted in a 26.5% increase in inventory turnover and an 18.2% reduction in transportation costs, highlighting the framework’s ability to achieve measurable economic and environmental benefits. Quantitative validation through rigorous methods, including RMSE and cost minimization calculations, confirms the framework’s capacity to address complex supply chain challenges. However, reliance on high-quality datasets and the absence of external variables such as geopolitical risks and market fluctuations are noted limitations. Exploring these gaps in future research will significantly improve the framework's scalability and practical implementation, especially in dynamic and multi-regional contexts. This study offers valuable insights into the integration of advanced analytics and optimization techniques, providing a practical foundation for developing resilient and sustainable supply chains in an increasingly unpredictable global context.