Skip to main content

Table 4 Graph-based clustering via Markov Stability performs equally well in a totally unsupervised manner, even if the number of clusters is not given a priori

From: Graph-based data clustering via multiscale community detection

DatasetMarkov Stability (c fixed to ground truth c)Markov Stability (c found by MS unsupervised)
Normalised Mutual Information (NMI)
Iris0.7980 (c=3)0.7980 (c=3)
Glass0.3932 (c=6)0.3687 (c=5)
Wine0.8347 (c=3)0.8347 (c=3)
WBDC0.7231 (c=2)0.7231 (c=2)
Control chart0.8489 (c=6)0.8520 (c=4)
Parkinson0.1334 (c=2)0.2973 (c=3)
Vertebral0.5971 (c=3)0.6018 (c=2)
Breast tissue0.5291 (c=6)0.5104 (c=3)
Seeds0.7142 (c=3)0.7142 (c=3)
Image Seg.0.5816 (c=7)0.6338 (c=13)
Yeast0.2909 (c=10)0.2909 (c=10)
Average0.58580.6023
Adjusted Rand Index (ARI)
Iris0.7455 (c=3)0.7455 (c=3)
Glass0.2380 (c=6)0.2086 (c=5)
Wine0.8471 (c=3)0.8471 (c=3)
WBDC0.8244 (c=2)0.8244 (c=2)
Control chart0.7056 (c=6)0.6824 (c=4)
Parkinson0.2659 (c=2)0.2667 (c=3)
Vertebral0.6096 (c=3)0.6113 (c=2)
Breast tissue0.3659 (c=6)0.3764 (c=3)
Seeds0.7432 (c=3)0.7432 (c=3)
Image Seg.0.4150 (c=7)0.4516 (c=13)
Yeast0.1746 (c=10)0.1746 (c=10)
Average0.53950.5393
  1. The best performance for each dataset is indicated in boldface