Skip to main content
. 2022 Sep 8;23:369. doi: 10.1186/s12859-022-04902-9

Table 2.

KluDo’s performance over the test datasets

LED MD MED RL
KK SP KK SP KK SP KK SP
Benchmark_1
OL 89.9 91.2 90.1 90.3 89.7 90.9 88.6 89.7
ARI 92.5 93.8 92.5 92.5 92.2 93.8 92.0 92.5
Benchmark_2
OL 77.6 79.5 76.9 77.6 78.2 79.5 74.4 74.4
ARI 85.3 85.9 85.9 85.3 85.9 85.3 84.0 82.1
Benchmark_3
OL 80.7 83.7 80.7 82.2 81.5 83.7 77.0 79.3
ARI 87.4 88.9 87.4 88.9 87.4 88.1 85.2 85.2
Islam
OL 88.0 89.3 86.7 86.7 89.3 88.0 82.7 82.7
ARI 92.0 93.3 90.7 93.3 90.7 93.3 90.7 92.0
Jones
OL 94.5 92.7 89.1 92.7 90.9 92.7 90.9 92.7
ARI 96.4 98.2 96.4 98.2 94.5 98.2 98.2 98.2
ASTRAL40
OL 84.0 84.7 84.0 84.7 84.1 84.7 83.2 83.8
ARI 87.3 87.8 87.4 87.1 87.4 87.7 86.9 86.9

KluDo’s accuracy for all combinations of the four kernels (LED, MD, MED and RL) and two clustering algorithms (kernel k-means and spectral clustering denoted by KK and SP, respectively) against the datasets Benchmark_1, Benchmark_2, Benchmark_3, Islam, Jones and ASTRAL40. The accuracies are based on the OL and ARI scores with the thresholds of 85% and 50%, respectively. The maximum accuracy in each row is illustrated in bold