Fig. 5

Feature selection and SVM model. A LASSO coefficient profiles of candidate DEIGs. The LASSO was used for regression of high dimensional predictors. The method uses an L1 penalty to shrink some unimportant regression coefficients to exactly zero. B Cross-validation to select the optimal tuning parameter log (Lambda) in LASSO regression analysis. The left dotted line represents the minimum error of the model, while the right dotted line represents the minimum number of features of the model within the range of error tolerance. In our study, we chose the model with the minimum error. C Performance measure of the cross-validated RF-RFE technique. When there are 23 features(variables), the model has the highest accuracy. D Features’ importance ranking in XGBoost analysis, a longer bar represented the variables had more influence on the outcome variables. E Venn diagram showing the hub genes shared by LASSO, RF-RFE, and XGBoost algorithm. F The ROC curve of the internal validation using SVM model