Table 2.

Classification rates produced by a CVA after each of the three dimensionality reducing approaches considered

		40 PC axes		PLS		Variable PC axes
Data acquisition	Data processing	RS rate	CV rate	RS rate	CV rate	# of axes used	RS rate	CV rate
Curve Tracing	Bending energy	100	69.6	76.1	73.9	9	82.6	87.0
Curve Tracing	Perpendicular projection	100	69.6	87.0	73.9	13	89.1	84.8
Curve Tracing	Elliptical Fourier analysis	100	58.7	87.0	65.2	12	87.0	73.9
Curve Tracing	Eigenshape analysis	100	65.2	82.6	54.4	24	93.5	69.6
Fan	Bending energy	100	65.2	89.1	82.6	7	89.1	89.1
Fan	Perpendicular projection	100	78.3	89.1	82.6	7	93.5	89.1

In each case, the canonical variate axes were significant using a Wilk's lambda test at p < 0.005 or better. Resubstitution (RS) were higher than cross-validation (CV) rates for the fixed number of principal component (PC) axes and partial least squares (PLS) methods, and closer to equal for the variable PC axes method. All rates are listed as percentages.