Unsupervised Clustering

The movement of participants between clusters suggests behavioural flexibility and possible responsiveness to intervention. Ward hierarchical clustering was selected over K-Means because it was considered more robust for heterogeneous small datasets. Ward hierarchical clustering was more stable than K-Means in the small real-world cohort. Among participants with complete paired data, 45% moved between clusters and 55% remained stable. The best clustering solution identified two stable self-management profiles at baseline and follow-up. Cluster-number selection combined silhouette scores with dendrogram inspection. The study retained k = 3 for macro-level monitoring and k = 5 for finer etiological differentiation.