Skip to main content
. 2022 Mar 11;9:839856. doi: 10.3389/fmed.2022.839856

FIGURE 2.

FIGURE 2

Predictive variables selection by the least absolute shrinkage and selection operator (LASSO) algorithm and random forest (RF) algorithm. (A) Identification of the optimal penalization estimate of lambda. Tenfold cross-validation with a minimum error criterion was used to determine the optimal penalization estimate of lambda in the LASSO regression. (B) The LASSO estimate profile of predictive variables. The left vertical line indicates the optimal lambda location, and the right vertical line indicates 1 standard error of optimal lambda. (C) Error convergence curve of the random forest model (500 trees). Optimal number of trees (n = 185) was determined according to the minimal error rate of fivefold cross-validation. (D) Importance ranking of candidate predictors in the random forest model. Mean decrease Gini index was calculated to define the importance of candidate predictors. (E) Venn diagram to determine identical predictors from LASSO and RF algorithm. A total of 7 identical predictors were determined from LASSO algorithm (8 variables) and RF algorithm (top 10 variables).