Table 5. Final selection of trained machine learning algorithms for time-window length of 10 seconds.

Classifier	Features	Approach	# Levels	Accuracy [%]	THR [%]	TMR [%]	TLR [%]	Kappa
Bagged Trees	HRNmean, HRstd, EDANmean, EDANFD, EDAnOR, EDAmmOR	SB	2	89.8	82.4	-	94.7	0.78
Bagged Trees	HRNmean, HRstd, EDANmean, EDANFD, EDAnOR, EDAmmOR	HR1	2	89.0	88.4	-	89.6	0.78
Bagged Trees	HRNmean, HRstd, EDANmean, EDANFD, EDAnOR, EDAmmOR	EDA2	2	85.3	86.0	-	84.3	0.69
Bagged Trees	HRNmean, HRstd, EDANmean, EDANFD, EDAnOR, EDAmmOR	SB	3	74.4	60.0	60.7	93.1	0.59
Bagged Trees	HRNmean, HRstd, EDANmean, EDANFD, EDAnOR, EDAmmOR	HR2	3	72.3	71.1	71.6	75.2	0.58

Results for two-level classification are shaded light gray.

THR, true high rate; TMR, true medium rate; TLR, true low rate.