Automatic design of decision-tree induction algorithms tailored to flexible-receptor docking data

BMC Bioinformatics

Table 3 Results of Experiment 2

Measure	Ligand	HEAD	C4.5	$\hat{μ}$	${\hat{σ}}^{2}$	t
Accuracy	ETH	0.70 ± 0.02	0.62 ± 0.02	0.08	0.00074	6.28
	PIF	0.87 ± 0.00	0.80 ± 0.02	0.06	0.00045	6.39
	TCL	0.65 ± 0.02	0.57 ± 0.02	0.07	0.00022	10.85
	NADH	0.75 ± 0.03	0.72 ± 0.02	0.03	0.00054	2.28
	INH	0.83 ± 0.02	0.79 ± 0.01	0.04	0.00033	4.57
	JPM	0.72 ± 0.02	0.65 ± 0.02	0.07	0.00061	5.90
Tree Size	ETH	30.20 ± 38.09	539.80 ± 29.64	510.0	2147.38	23.95
	PIF	17.80 ± 13.44	283.00 ± 39.97	265.2	1174.40	16.84
	TCL	43.00 ± 37.49	588.00 ± 37.39	545.0	3193.11	20.99
	NADH	87.00 ± 36.42	360.00 ± 22.33	273.0	2379.78	12.18
	INH	42.80 ± 27.96	278.60 ± 24.80	235.8	1313.29	14.16
	JPM	121.80 ± 65.60	483.80 ± 19.76	362.0	4152.00	12.23

Table3 presents the classification accuracy and tree size of HEAD-DT and C4.5 in the ETH, PIF, TCL, NADH, INH, and JPM data sets. HEAD-DT results were obtained by running an evolved decision-tree algorithm tailored to each data set.

ISSN: 1471-2105