Embedded Files
The experimental investigation into clustering methodologies applied to the POJ104 dataset reveals nuanced patterns in algorithmic performance and cluster number determination. Across K-means, Agglomerative, and GMM approaches, the internal validation metrics exhibit characteristic trade-offs between cluster cohesion and separation.
Finding 1: Agglomerative clustering demonstrates marginal advantages in specific metrics, such as DBI and SSE, after certain thresholds (k = 247 and k = 57, respectively). However, all methods exhibit context-dependent trade-offs, with no single method consistently outperforming the others across all evaluation metrics.
Finding 2: The proposed algorithm achieves strong performance, with a normalized mutual information (NMI) score of 79%, surpassing the 0.5 threshold for meaningful cluster separation. Despite ARI’s low values (< 0.08), the algorithm proves effective for unsupervised tasks, particularly in complex distributions lacking ground-truth labels.
Page updated
Google Sites
Report abuse