Fig. 3: The audiometric factors associated with SFHL in the training sets.
From: Prediction of risk of hearing loss by industry noise from cross-sectional and longitudinal data
A, B LASSO coefficient profiles. Shown are the trajectories of the coefficients (y-axis) for candidate audiometric variables across a sequence of regularization penalty values (log(λ), top x-axis). The vertical dashed line indicates the λ value selected via 10-fold cross-validation, which yielded the most parsimonious model. The shaded bands around each coefficient path represent the variability (standard error) of the coefficient estimates across the cross-validation folds. The analyses are descriptive of variable selection and no inferential statistical tests or P values are generated at this stage. Panels are shown for (A) male (n = 4310 biologically independent participants) and (B) female (n = 743 biologically independent participants) workers. C, D Random Forest feature importance ranking. Variables are ranked by their mean decrease in Gini impurity (x-axis), a measure of their contribution to classifying SFHL in the Random Forest ensemble. Error bars represent the standard deviation of the importance measure across all trees in the forest. Panels are shown for (C) male (n = 4,310) and (D) female (n = 743) workers.
