Skip to main content

Table 2 Comparison of graph constructions in terms of clustering performance (NMI and ARI) on eleven UCI datasets

From: Graph-based data clustering via multiscale community detection

DatasetPMSTε-ballRMST k=1kNNCkNN δ=1
   γ=0.5γ=0.25γ=0.125k=3k=7k=12k=3k=7k=12
Normalised Mutual Information (NMI)
Iris0.77640.80570.71060.77640.74000.76270.80570.80570.82260.79800.7777
Glass0.38860.38620.36560.36460.39530.38630.39270.36260.35170.39410.4170
Wine0.80800.79720.73890.84000.86330.79550.83360.82150.75280.83470.8113
WBDC0.60420.58190.58390.60420.61820.61880.59440.61210.58220.72310.6113
Control chart0.82720.84040.76020.71330.81300.82660.85200.8520.80780.88830.8531
Parkinson0.22200.20650.20120.21430.28110.31750.28150.21760.21130.29730.2423
Vertebral0.50430.62730.54320.53230.59990.60600.59280.54240.64500.60180.6116
Breast tissue0.57220.56160.52980.58200.54610.54470.55250.52430.52530.56480.5601
Seeds0.72930.69430.73610.75700.68100.75150.74580.73180.63840.73810.7245
Image Seg.0.69480.66050.67620.70720.74880.61540.64650.66670.60120.63470.6541
Yeast0.28810.30510.29520.26260.25630.27640.29970.30800.24730.29590.3072
Average0.58350.58790.55830.57760.59450.59100.59970.58590.56230.61550.5973
Adjusted Rand Index (ARI)
Iris0.74200.75920.66030.74200.69570.71910.75920.75920.81840.74550.7445
Glass0.23230.20990.19830.20290.22580.22780.22310.22660.21340.23980.2496
Wine0.83500.80720.73750.87120.88230.80250.84980.83490.74140.84710.8360
WBDC0.71930.71140.70140.71930.73690.73680.70100.73100.66970.82440.7200
Control chart0.69290.73640.56940.53710.69910.67480.68240.70710.69020.82800.7227
Parkinson0.22670.22050.20380.15400.25560.21760.20010.20450.11650.26670.2101
Vertebral0.52570.64450.57020.54110.60150.59820.58020.53300.64410.61130.6302
Breast tissue0.46890.44940.41000.40170.34940.40120.42720.40780.36310.37640.4471
Seeds0.73530.74970.74580.75890.66870.78760.78890.77610.64020.77640.7655
Image Seg.0.60600.51440.60300.61930.59420.38000.46690.54710.37910.45220.5121
Yeast0.19080.23680.18270.17720.16490.17550.22300.25310.17970.19420.2294
Average0.54320.54900.50750.52040.53400.52010.53650.54370.49600.56020.5516
  1. A high NMI (and ARI) indicate that the best partition found by Markov Stability is similar to the ground truth, i.e., better clustering. The best performance for each dataset is marked in bold