Optimizing Purchase Predictions in Retail: A Network Science and Artificial Intelligence Approach

Abstract

Accurately predicting customer purchase behavior is essential for optimizing inventory management and marketing strategies in the retail sector. This study proposes a novel integration of network science measures and artificial intelligence techniques to significantly enhance the accuracy of purchase predictions. Specifically, we constructed a co-purchase network from the Instacart dataset and integrated centrality measures-such as degree centrality, closeness centrality, betweenness centrality, and eigenvector centrality-as new informative features into machine learning models including Random Forest (RF), Gradient Boosting (GB), and Support Vector Machine (SVM). Our approach notably improved prediction accuracy, with the Random Forest model achieving the highest Area Under the Curve (AUC) of 0.82. These findings demonstrate the potential of network-based features to uncover hidden patterns in consumer behavior, providing retailers with robust predictive tools and enhancing operational efficiency and customer satisfaction.