Enhancing E-Commerce Innovation with Predictive Analytics in Business Management Using MARL-GNN Models

The rapid growth of e-commerce has increased the demand for comprehensive predictive analytics to improve commercial decision-making. Conventional predictive models, such Logistic Regression and Random Forest, fail to accurately represent the complex relationships between consumers and items, hence restricting their accuracy and flexibility. This study introduces MARL-GNN, an innovative framework that integrates Multi-Agent Reinforcement Learning (MARL) with Graph Neural Networks (GNN) to represent dynamic customer-product interactions. MARL agents enhance decision-making, whereas GNNs effectively reflect transactional data. The suggested method significantly surpasses current models, with an accuracy of 96.4%, precision of 94.7%, recall of 95.2%, and a reduced MAE of 0.023. In comparison to traditional methods, MARL-GNN effectively adjusts to changing market trends and consumer behaviours, delivering enhanced demand forecasting and business optimisation. The results validate the effectiveness of graph-based reinforcement learning for instantaneous decision-making in e-commerce. Future study may investigate real-time implementation and enhanced explainability to facilitate more industry usage.