F-Ensemble: A Full Fusion Ensemble Model for Predicting E-commerce User Purchasing Behavior

Mining historical behavioral data of e-commerce users to help platforms identify users with the highest potential consumption value is a hot topic in the e-commerce field. However, current research has not adequately considered the impact of imbalanced user purchasing behavior data on model prediction. The structure of existing prediction models is relatively simple, the user behavior features involved are not comprehensive enough, and the prediction accuracy needs to be improved. Therefore, this paper proposes an F-Ensemble model based on a full fusion ensemble learning strategy to predict the purchasing behavior of users. This model uses decision tree, multilayer perceptron, random forest, and AdaBoost as base models, integrating both single models and ensemble learning models to predict the purchasing behavior of users. The experimental results show that the Macro-F1 score of the F-Ensemble model based on the full fusion ensemble learning strategy reaches 0.86485, representing a 5.30% improvement over the worst-performing decision tree model. Additionally, the F-Ensemble model significantly improves the F1 score of purchase behavior prediction under imbalanced data conditions, achieving a score of 0.73283, which is 13.39% higher than the performance of the worst-performing decision tree model. This result further validates the effectiveness of the F-Ensemble model. In conclusion, the F-Ensemble model constructed in this paper enhances the prediction accuracy of e-commerce users' purchasing behavior and holds practical value by reducing platform management costs.