Table 2. Performance of three different implementations of Linear Discriminant Analysis.

Data Set		Misclassification error (%)
	Episode type	myLDA	dnldLDA	MATLAB's classify
Training Set	Total	46.13	29.80	44.17
	0–2 days	31.88	2.06	34.14
	3–6 days	74.80	76.57	73.08
	7–9 days	72.03	84.22	60.72
	> = 10 days	71.96	89.75	51.67
Test Set	Total	47.20	32.32	47.13
	0–2 days	32.46	2.72	35.85
	3–6 days	75.62	82.46	71.31
	7–9 days	75.41	90.75	65.78
	> = 10 days	75.09	92.16	67.35

Per class misclassification error was lower in Class A (0–2 days) than in classes B (3–6 days), C (7–9 days) and D (> = 10 days). The dnldLDA routine was >97% accurate in predicting Class A. Misclassification errors in classes B, C, and D were high for all three routines, re-emphasizing the overlap shown in Figure 6 .