Shannon Entropy

Shannon entropy biomarkers achieved the strongest non-nested cross-validated performance among the tested feature families. Shannon entropy was interpreted as more discriminative than sample entropy or Higuchi fractal dimension because it captures distributional variability across utterances. Permutation testing found the entropy model statistically significant relative to shuffled labels. Shannon entropy biomarkers quantified distributional uncertainty in acoustic trajectories. The strongest entropy markers were mainly acoustic variability entropy features rather than raw acoustic level features.