Table 1.

Training and test performance (mean ± standard deviation) comparisons of three methods of feature selection (best performing single feature, significant between-group, and stepwise optimization) in ${\mathrm{O}}_2\mathrm{Hb}$ and HHb connectivities.

	FC (entire measurement)		FC (baseline)		FC (task)
	${\mathrm{O}}_2\mathrm{Hb}$	HHb	${\mathrm{O}}_2\mathrm{Hb}$	HHb	${\mathrm{O}}_2\mathrm{Hb}$	HHb
Training subsets
Single feature	$77.1\pm 1.7\%$	$80.5\pm 3.5\%$	$76.5\pm 3.0\%$	$79.7\pm 6.1\%$	$81.4\pm 1.7\%$	$83.4\pm 2.9\%$
Significant features	$78.0\pm 1.3\%$	$78.0\pm 3.0\%$	$78.0\pm 1.8\%$	$78.4\pm 1.8\%$	$79.7\pm 3.0\%$	$78.0\pm 2.6\%$
Stepwise selection	$84.7\pm 1.0\%$	$89.8\pm 1.1\%$	$86.5\pm 2.5\%$	$88.2\pm 3.3\%$	$89.7\pm 2.6\%$	$89.9\pm 2.5\%$
	$F_{(\mathrm{2,12})}=48.5$$p<0.001$	$F_{(\mathrm{2,12})}=25.9$$p<0.001$	$F_{(\mathrm{2,12})}=24.0$$p<0.001$	$F_{(\mathrm{2,12})}=8.28$$p<0.01$	$F_{(\mathrm{2,12})}=23.5$$p<0.001$	$F_{(\mathrm{2,12})}=24.6$$p<0.001$
Test subsets
Single feature	$73.9\pm 13.5\%$	$76.3\pm 14.0\%$	$75.2\pm 8.4\%$	$71.4\pm 27.2\%$	$78.5\pm 10.4\%$	$80.3\pm 21.1\%$
Significant features	$76.5\pm 4.9\%$	$77.0\pm 9.5\%$	$72.4\pm 8.7\%$	$71.5\pm 9.5\%$	$75.2\pm 15.8\%$	$77.0\pm 9.5\%$
Stepwise selection	$82.8\pm 2.2\%$	$88.2\pm 4.0\%$	$85.6\pm 7.9\%$	$89.2\pm 11.7\%$	$90.4\pm 9.6\%$	$90.6\pm 9.6\%$
	$F_{(\mathrm{2,12})}=1.49$$p>0.05$	$F_{(\mathrm{2,12})}=2.24$$p>0.05$	$F_{(\mathrm{2,12})}=3.50$$p>0.05$	$F_{(\mathrm{2,12})}=1.63$$p>0.05$	$F_{(\mathrm{2,12})}=2.13$$p>0.05$	$F_{(\mathrm{2,12})}=1.20$$p>0.05$