Table 3 Comparative Evaluation of Five Non-negative Matrix Factorization Algorithms for Clustering with 2 Clusters. Source data are provided as a Source Data file

From: Unbiased clustering of acute-on-chronic liver failure patients using machine learning in a real-world ICU cohort

Algorithms

K

PAC

RSS

EVAR

Sil

CCC

Disp

SC

Brunet

2

0.1168

33737.2772

0.7409

1.0000

0.9858

0.9058

0.9531

Lee

2

0.0000

33190.9718

0.7451

1.0000

1.0000

1.0000

1.0000

nsNMF

2

0.0039

33712.2085

0.7411

1.0000

0.9997

0.9968

0.9986

snmf/r

2

0.0000

33191.8259

0.7451

1.0000

1.0000

1.0000

1.0000

snmf/l

2

0.0000

33191.8224

0.7451

1.0000

1.0000

1.0000

1.0000

  1. The smaller the PAC, RSS and Disp values are, the better the clustering effect is. The larger the EVAR, Sil, CCC, and SC values are, the better the clustering effect is.
  2. PAC: the partitioning around medoids with ambiguous clustering; RSS: residual sum of squares; EVAR: explained variance; Sil: silhouette coefficient; CCC: cophenetic correlation coefficient; Disp: dispersion; SC: silhouette consensus.
Back to article page