Enhancing Model Interpretability Through Interactive Visual Analysis and Counterfactual Explanation Methods

Counterfactual explanation is a post-modeling technique that helps users manipulate input features to achieve desired model decisions. This study employs counterfactual explanations to help users understand predictive outcomes and their underlying reasons. By reversing model decisions, user issues can be addressed. We developed a visual analysis framework combining machine learning algorithms and visual analytics. And our primary efforts are outlined as follows: Establish a visual analysis framework combining machine learning algorithms and visual analysis. Use counterfactual interpretation, we improved model interpretability and helped users understand prediction results. Design visualization views according to the visual analysis tasks derived from user needs based on machine learning models, and designed counterfactual explanation operation views for model decision instances. Integrate the visual analysis view, realize the interactive visual analysis system CFEVis based on the credit approval data prediction model.