Free Access

Table 3.

Performance of different algorithm and parameter combinations on the 100 main OCs.

Algorithm Parameters TP FP TN FN Sensitivity Specificity Precision
DBSCAN ϵACG, 1 repeat 22 0 60 18 0.55 1.00 1.000pt0.35cm
ϵACG, 30 repeats 20 0 60 20 0.50 1.00 1.00
ϵc 20 1 59 20 0.50 0.98 0.95
ϵn1 20 0 60 20 0.50 1.00 1.00
ϵn2 7 2 58 33 0.17 0.97 0.78
ϵn3 2 0 60 38 0.05 1.00 1.00
{ϵc,  ϵn1,  ϵn2,  ϵn3} 25 2 58 15 0.62 0.97 0.93

HDBSCAN mclSize = 80 17 3 57 23 0.42 0.95 0.850pt0.35cm
mclSize = 40 24 6 54 16 0.60 0.90 0.80
mclSize = 20 31 7 53 9 0.78 0.88 0.82
mclSize = 10 33 7 53 7 0.82 0.88 0.82

GMM ms = 800 7 0 60 33 0.17 1.00 1.000pt0.35cm
ms= variable 13 0 60 27 0.33 1.00 1.00

Notes. True positive (TP), false positive (FP), true negative (TN) and false negative (FN) counts of detected clusters are given along with the sensitivity, specificity and precision. 68.3% confidence intervals are shown for all numbers. Confidence intervals for a handful of values (e.g. measured precisions of exactly 0.0 or 1.0) were adjusted to include the measured values. This corrects for approximations in the calculation of binomial confidence intervals where the measured success probability is exactly 0 or 1. All objects that did not pass the criterion in Sect. 4.2 with a CST greater than 3σ were discarded before crossmatching and producing this table.

Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.

Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.

Initial download of the metrics may take a while.