Feature-Space Separability

K-mer features increased dispersion but did not improve separability. PCA projections of composition, k-mer, and GA-selected k-mer features showed no clear class separation. Clustering failed to recover Parkinson class labels. High Silhouette values from length-based clustering reflected internal structure unrelated to Parkinson labels.