Biomedical Engineering Reference
In-Depth Information
CNN - Middle (192ms)
RNN - Middle (192ms)
CNN - Large (500ms)
RNN - Large (500ms)
2
1
0
5
10
15
20
25
30
35
40
Class 1 (40 Patients)
2
1
0
5
10
15
20
25
Class 2 (27 Patients)
2
1
0
2
4
6
8
10
12
Class 3 (13 Patients)
2
1
0
5
10
15
20
25
Class 4 (29 Patients)
2
1
0
2
4
6
8
10
12
14
16
18
20
Class 5 (21 Patients)
Fig. 5.6 Error performance with different prediction time horizons (CNN vs. RNN) Here, the
average NRMSEs for CNN and RNN are 0.16 and 0.33, respectively
numbers 8 and 86) and three in class 2 (patient numbers 22, 38 and 51). We also
show the average error performance for each class in Table
5.3
. In the short time
prediction (38 ms), all the classes have improved more than 30 %. The 50 %
improvement was achieved in classes 3, 4, and 5 of the large time prediction
(500 ms). As shown in prediction error of Table
5.3
, the proposed CNN works for
any five classes, thus there are no particular differences of error among the five
classes because the criterion of feature selections in CNN is designed to minimize
the error.
To compare the experimental results with other peer studies, we used experi-
mental results of (i) optimized adaptive neural network prediction (O-ANN) [
31
]
Table 5.3 Average error performance among a variety of prediction time (CNN vs. RNN)
Prediction time horizon (CNN/RNN)
38.46 ms
115.38 ms
192.3 ms
269.23 ms
346.15 ms
423.07 ms
500 ms
Class
1
0.088/
0.262
0.104/
0.299
0.121/
0.344
0.137/
0.388
0.157/
0.455
0.179/
0.551
0.222/
0.766
Class
2
0.089/
0.260
0.109/
0.349
0.130/
0.430
0.150/
0.510
0.171/
0.588
0.198/
0.708
0.237/
0.991
Class
3
0.144/
0.491
0.160/
0.541
0.177/
0.617
0.192/
0.675
0.214/
0.738
0.255/
0.863
0.314/
1.012
Class
4
0.125/
0.354
0.139/
0.387
0.156/
0.472
0.173/
0.524
0.191/
0.610
0.220/
0.701
0.274/
0.847
Class
5
0.098/
0.294
0.110/
0.440
0.125/
0.495
0.145/
0.558
0.175/
0.625
0.208/
0.708
0.260/
0.815
(Unit: NRMSE)
Search WWH ::
Custom Search