Figure 3
From: Model selection to achieve reproducible associations between resting state EEG features and autism
Model predictivity (quantified by area under the receiver operating characteristic, or “AUROC”), coefficient reproducibility (quantified by Spearman’s rank correlation coefficient), and overall model robustness as a function of L2-regularized logistic regression regularization strength. The model (logistic regression, or “LR,” + scale invariant residuals transform, or “SIRT,” preprocessing) corresponding to the highest achieved robustness score is highlighted by the vertical magenta band. The model corresponding to the highest achieved predictivity is highlighted by the vertical red band. \(\:x\)-axis values represent increasing L2 regularization strength (from left to right), corresponding to the negative \(\:{\text{log}}_{10}\)-transformed regularization hyperparameter values used in scikit-learn’s implementation of L2-regularized logistic regression.
