Table 2.

Comparison of SUMU, SUMU-LDA, the three variants of osort: osort1, osort3, osort_C, and the three variants of the isolation score: isol75, isol50, isol10 in evaluating the quality of separation on synthetic data provided with osort (Rutishauser et al 2006). The table is organized similarly to table 1. No classification is significantly higher than the naive algorithm

Algorithm	SUMU	osort1	osort3	osort_C
Accuracy	80.8 ± 6.1	57.1 ± 13.2	53.1 ± 15.9	57.1 ± 13.2
False-positive rate	12.0 ± 8.9	42.9 ± 13.2	46.9 ± 15.9	42.9 ± 13.2
False-negative rate	7.2 ± 3.9	0.0 ± 0.0	0.0 ± 0.0	0.0 ± 0.0

Algorithm	SUMU-LDA	isol75	isol50	isol10
Accuracy	75.0 ± 7.3 (79.5 ± 0.6)	75.7 ± 10.0	80.8 ± 6.1	67.7 ± 8.8
False-positive rate	15.3 ± 5.9 (13.6 ± 0.5)	5.9 ± 5.9	16.5 ± 8.7	32.4 ± 8.8
False-negative rate	9.7 ± 5.2 (6.8 ± 0.3)	18.4 ± 5.9	2.8 ± 2.8	0.0 ± 0.0