TriFusion Ensemble Model: A Physical Systems Approach to Enhancing E-Commerce Predictive Analytics with an Interpretable Hybrid Ensemble Using SHAP Explainable AI

In the rapidly evolving e-commerce landscape, predicting consumer purchasing behavior is crucial for optimizing marketing strategies and improving sales. Traditional machine learning models often deliver high accuracy but lack interpretability, limiting their real-world application. This paper introduces the TriFusion Ensemble Model, a novel framework that not only combines Random Forest, Support Vector Machine (SVM), and Logistic Regression but also uniquely incorporates concepts from physical systems. In this approach, each base model is treated as a force contributing to the overall prediction, while the meta-learner acts as a balancing mechanism, minimizing classification error akin to reducing the system’s energy. This novel borrowing from physical systems is groundbreaking, as it reframes the model’s optimization process, enhancing both predictive accuracy and transparency. By structuring model interactions through energy minimization, the TriFusion Ensemble addresses the black-box nature of advanced machine learning models. Evaluated using AUC-ROC and SHAP-based feature importance analysis, the model shows moderate performance improvements over individual models while enhancing feature transparency. This innovative approach can extend beyond e-commerce to fields like healthcare and finance, offering a new paradigm for deploying complex machine learning models with both accountability and practical applicability, grounded in physical system principles.