Interpretable Coronary Artery Disease Prediction Based on K-Stacking Model

To improve the accuracy of ensemble learning methods in predicting coronary artery disease and to increase the interpretability of the model. A modified weighted stacking model called the K-Stacking model has been proposed. Based on the stacking model, the Kappa coefficient is considered in the weight to fully capture the effective information in the base model. And performed a local and global interpretability analysis of the model by introducing SHAP techniques to assess the importance of features in predicting the output. Compared with the traditional stacking model and a single classification model, the evaluation indicators of the K-Stacking model are significantly improved. The K-Stacking model can achieve up to 97.82% accuracy, 98.52% recall and 97.09% precision, respectively. The use of SHAP technology for interpretability analysis also improves the confidence of the model prediction results.