Relationship Matrix Nonnegative Decomposition for Clustering

Table 4

Clustering accuracy and normalized mutual information of 𝐾-means, spectral clustering (SpeClus), SNMF, WNMF, and RMND on the USPS dataset.

Cluster number	Clustering accuracy					Normalized mutual information
𝑘	𝐾-means	SpeClus	SNMF	WNMF	RMND	𝐾-means	SpeClus	SNMF	WNMF	RMND
2	0.9650	0.9650	0.9650	0.9660	0.9900	0.8135	0.8135	0.9650	0.9660	0.9900
3	0.8867	0.8467	0.8700	0.8402	0.8912	0.6917	0.6380	0.6689	0.6656	0.7189
4	0.8015	0.7850	0.8057	0.7934	0.8421	0.6430	0.5904	0.6012	0.6118	0.6634
5	0.8392	0.7700	0.7864	0.7686	0.8354	0.6826	0.6147	0.6252	0.6195	0.6586
6	0.7330	0.7300	0.7206	0.6928	0.7325	0.6140	0.6110	0.5921	0.5633	0.6061
7	0.7194	0.7186	0.7155	0.6787	0.7331	0.6041	0.6063	0.5925	0.5656	0.6116
8	0.7199	0.7302	0.7194	0.7025	0.7463	0.6239	0.6320	0.6232	0.5987	0.6494
9	0.6932	0.7358	0.6937	0.6702	0.7007	0.5902	0.6406	0.5910	0.5640	0.5922
10	0.6343	0.6860	0.6527	0.6216	0.6725	0.5556	0.6023	0.5567	0.5333	0.5605
ave.	0.7769	0.7741	0.7699	0.7482	0.7938	0.6465	0.6388	0.6462	0.6320	0.6723