Identifying risk factors for vasculogenic etiology in patients with erectile dysfunction based on clinical features and machine learning

Figures

• 

  Figure 1. Study flowchart.

• 

  Figure 2. Boruta feature importance plot. The output of the Boruta algorithm is shown in the figure, where the blue lines represent the shadow scores (minimum, mean, and maximum). Variables are classified according to the color of the boxplots: green for important variables, yellow for tentative variables, and red for rejected variables.

• 

  Figure 3. LASSO cross - validation error plot. The relationship between the λ value and binomial deviance is shown in the figure. Two vertical dashed lines in the figure mark two key positions: the left dashed line corresponds to the minimum mean squared error, and the right dashed line corresponds to the position of the minimum mean squared error plus one standard error.

• 

  Figure 4. LASSO coefficient path plot. The figure illustrates the relationship between the L1 norm (i.e., the regularization term of the model) and the individual regression coefficients in the LASSO regression.

• 

  Figure 5. ROC curves of multiple models. The ROC analysis results of various machine learning models for vascular ED are shown in the figure. Among them, AUC is the area under the curve; LR is logistic regression; RF is random forest; SVC is support vector classification; LGBM is light gradient boosting machine; XGBM is extreme gradient boosting machine.

• 

  Figure 6. SHAP feature importance bar plot. The figure displays the SHAP importance of seven features in the logistic regression model in descending order, measured by the mean absolute SHAP value of each feature to reflect its impact on the model’s prediction. A longer bar indicates a greater contribution and higher importance of the variable.

• 

  Figure 7. SHAP beeswarm plot. Each dot represents an individual sample. The horizontal axis represents the SHAP value, reflecting the feature’s impact on the model output: positive values increase the predicted risk, while negative values decrease it. The color gradient (from blue to red) corresponds to the feature value, ranging from low to high. Features are ordered by the magnitude of their average SHAP value, reflecting their overall importance to the model.

• 

  Figure 8. SHAP dependence plot combination. The x-axis represents feature values, and the y-axis represents SHAP values (impact on model output). Blue points represent individual samples, and the red line indicates the trend. Positive values increase the predicted risk, while negative values decrease it.

• 

  Figure 9. Scatter plot of SHAP contributions for multiple samples. Each point represents an individual sample. The horizontal axis is the SHAP value, where positive values increase the predicted risk and negative values decrease it. The color of each point indicates the direction of the contribution value, and its size reflects the magnitude of the absolute contribution. Features are listed in descending order of their impact on the model.