Skip to main content
. Author manuscript; available in PMC: 2019 Feb 1.
Published in final edited form as: Ann Biomed Eng. 2017 Nov 27;46(2):233–246. doi: 10.1007/s10439-017-1966-6

Table 4.

Optimal feature combinations for each algorithm that provide the best NRMSE (top) or CC (bottom). Results are given as mean ± standard deviation.

Algorithm Interval (s) FR Method FR Norm Channel Selection NRMSE CC Computational Time per Output
NARMA 0.5 Boxcar 1 Highcc 17 ± 7% 0.69 ± 0.35 2.2 ± 0.8 ms
LinReg 0.5 Boxcar 1 Highcc 22 ± 6% 0.51 ± 0.30 2.1 ± 0.6 ms
Kalman 2 Boxcar 1 Highcc 25 ± 15% 0.67 ± 0.40 3.4 ± 1.2 ms
Algorithm Interval (s) FR Method FR Norm Channel Selection NRMSE CC Computational Time per Output

NARMA 1 SmoothBin 0 Highcc 21 ± 8% 0.72 ± 0.33 8.0 ± 4.3 ms
LinReg 2 SmoothBin 1 Highcc 32 ± 34% 0.70 ± 0.29 9.0 ± 4.5 ms
Kalman 0.1 TriangleC 0 Lasso 27 ± 15% 0.81 ± 0.13 2.0 ± 0.6 ms