Figure 3.

Feature selection and performance of machine learning in ROI method. A, Mean square error (MSE) of each feature set in the ROI method. The minimum MSE was 287.5 when the first six features (according to absolute value of ${\beta }_i$) were used in the training set via feature selection. B, Correlation between actual and predictive VHI scores. Favorable and significant results were achieved with an r value of .7516 and an R² value of .5649, indicating that this model can predict severity of hypokinetic dysarthria. C, Distribution of R² via permutation test in ROI method. A significant P value (P < .001) was achieved via permutation test