Model Transparency: Integrating XGBoost with SHAP for Explainable Machine Learning

This paper examines integrating the XGBoost algorithm with SHAP values to balance predictive performance and model interpretability. XGBoost is widely recognized for its high accuracy and efficiency, yet its ensemble structure makes the internal decision-making process difficult to interpret. SHAP offers a theoretically grounded framework based on Shapley values that enables both global and local explanations of model behavior. The paper focuses on identifying key predictors using SHAP summary analysis, exploring variable interactions, and providing detailed explanations of individual predictions through local SHAP visualizations. The results show that combining XGBoost with SHAP creates a robust and transparent modeling framework suitable for domains where explainability is essential. Moreover, SHAP uncovers complex feature relationships that traditional feature-importance methods miss, thereby improving the overall interpretive value of the model.