Fig. 2

Results based on machine learning model output. A Evaluation of the four machine learning algorithms based on the area under the curve AUC of the ROC curve. B Importance matrix plot of the top 20 features selected based on LightGBM machine learning algorithms and SHAP value, showing the relative contribution of each variable to NAFLD. C SHAP summary plot of the top 20 features selected based on LightGBM machine learning algorithms and SHAP values, in which one dot per individual per feature is coloured in accordance with an attribution value, with red denoting a greater value and blue denoting a lower value. A higher SHAP value indicates a greater risk of NAFLD