Predictive Analytics for Supply Chain Optimization using ML Models

The increasing complexity and internationalization of recent supply chains have made the rule-based decision-making systems of the past no longer appropriate to address efficiency and resilience. It is suggested that a supply chain predictive analytics framework based on pagewise supply chain optimisation is developed on the basis of recent paradigm-shifting machine learning models that include Temporal Fusion Transformer (TFT), Graph Neural Networks (GNNs), TabNet, and NGBoost. It uses historical sales and logistics data as well as supplier networks to dynamically predict demand, highlight risks in the supply chain, and adapt inventory levels using advanced deep learning and probabilistic modeling techniques. TFT is designed for interpretable multi-horizon demand forecasting, GNNs capture relational dependencies among supply nodes, TabNet enables trustworthy decision making over structured tabular datasets, and NGBoost enables uncertaintyaware risk assessment. The part comes with a real-time dashboard visualisation and ERP connection, which means that data-driven decisions can be made at scale. The experimental evaluation demonstrates considerable improvements in prediction quality, cost and agility over the state- of-the-art approaches. This study presents a new, understandable and scalable AIbased supply chain decision support system that ushers in a new generation of predictive logistics.