Fig.3.

Enhancement of a FNN over a,c) SVM and b,d) LR in CNP. In a,b), data are enhanced with increasing λ (see Algorithm 2). Solid lines represent the accuracy for f₁ (FNN), while dashed lines show the accuracy for f₂ (SVM in a and LR in b). Error bars reflect standard deviation across 10 random seeds of k-fold cross-validation initialization seeds for FNN. Line color shows the suppression coefficient for f₂, η. In c,d), the correlation between original and enhanced features is shown with increasing λ. The original and enhanced features are highly correlated (r′s > 0.94).