Real-Time Data-Driven Decision Support in Retail: A Hybrid GraphSAGE+XGBoost Model for Predicting Reorder Behavior and Unraveling Consumer Communities

The rising demand for real-time, data-driven decision support in retail platforms has underscored the need for intelligent systems capable of modeling both behavioral sequences and product relationships. This study introduces a hybrid architecture for real-time decision support in retailing by coupling graph-based learning with conventional machine learning methods. Based on Instacart 2017 data, it constructs a heterogeneous user-product graph and utilizes GraphSAGE to obtain relational embeddings. This combination of embeddings and domain-specific features is then fed into an XGBoost classifier to predict reorder behavior. Empirical findings show that the proposed GraphSAGE+XGBoost model outperforms conventional baselines, including the sole XGBoost, Multilayer Perceptron (MLP), and Long Short-Term Memory (LSTM) models. In particular, the hybrid model outperformed all baselines across all metrics, achieving a precision of 0.82, a recall of 0.78, an F1-score of 0.76, and a mean Average Precision (mAP) of 0.75. Furthermore, within the co-purchase network, product-level community identification identified significant clusters (such as breakfast staples, health-conscious products, and impulsive snacking) that provided insights into customer demographics and marketing potential. The experimental analysis comparing the proposed GraphSAGE+XGBoost with baseline models, including LSTM, XGBoost, and MLP, demonstrates that the proposed hybrid model outperforms in terms of modeling accuracy, Precision, and generalizability. The system is optimized for real-time inference and can operate in a dynamic commercial landscape, unraveling complex co-purchase behavior and hidden consumer communities.