Table 2.

Results of parameter optimization for AR-based methods UAR, VAR, and BAR without the prediction error variance feature

	UAR			VAR			BAR
	p 0	p	log(UC)	p 0	p	log(UC)	p 0	p	q	log(UC)
A01	0.582	13	−2.8	0.612	4	−2.6	0.601	8	2, 2	−0.8
A02	0.446	6	−3.0	0.461	6	−2.8	0.461	14	1, 1	−0.6
A03	0.573	12	−2.6	0.625	2	−2.8	0.578	12	1, 3	−2.6
A04	0.418	10	−2.2	0.395	4	−2.2	0.421	12	2, 2	−2.6
A05	0.406	4	−2.6	0.410	2	−2.4	0.418	5	1, 2	−2.2
A06	0.429	15	−2.2	0.434	12	−2.2	0.457	15	1, 1	−2.6
A07	0.544	14	−2.6	0.533	13	−2.4	0.559	14	1, 3	−2.6
A08	0.635	15	−2.4	0.673	4	−2.4	0.639	5	1, 2	−2.4
A09	0.614	3	−2.2	0.640	3	−2.0	0.623	7	1, 2	−2.2
Global	0.494	13	−2.6	0.507	4	−2.6	0.499	13	1, 1	−2.4

All nine subjects (A01, A02, ) are shown. Columns show the 0.9 quantile of the classification accuracy p ₀, linear model order p, bilinear model order q, and update coefficient logUC. The last row shows the results of the global optimization