Table 2.

Predictive capacity of models as determined by cross-validation

Classification	Dataset	Validation method	0.50% peak shift window		4.0% peak shift window
			GA	SVM	GA	SVM
Air vs. EHC-93	dewarped	One out	78.41	70.08	79.17	66.67
		Random	65.83	60.83	71.51	73.33
		K-folds	73.86	61.36	70.83	70.83
	non-dewarped	One out	a	a	57.73	52.23
		Random	a	a	61.46	55.21
		K-folds	a	a	47.73	48.64
0 h vs. 24 h	dewarped	One out	72.83	81.82	66.67	87.50
		Random	72.92	76.04	58.33	87.50
		K-folds	72.73	72.73	81.44	83.58
	non-dewarped	One out	a	a	54.55	72.73
		Random	a	a	68.75	66.67
		K-folds	a	a	63.64	72.73

Reliability (future predictive capacity) of discriminatory models generated using dewarped and non-dewarped datasets as determined by cross validation using the One-out, the Random or the K-folds methods. Values provided are the percent averages of the predictive capacities of a number of models generated by different combinations of chromatograms as training (model generation) and test (model validation) data. GA and SVM were used in model generation. Cross validation was not possible when there was an insufficient number of recalibratable chromatograms within an exposure or time of recovery group, and is indicated by a 'a'.