Table 3.
The average relative error of BP neural network with different parameters.
| Subjects | Average Relative Error (ARE, %) | |||
|---|---|---|---|---|
| Individuals (n = 24) | Binary Mixtures (n = 75) | Ternary Mixtures (n = 51) | Mean Value | |
| a. Training functions | ||||
| trainlm | 5.0 | 10.0 | 19.4 | 11.5 |
| traingd | 6.0 | 10.1 | 21.0 | 12.4 |
| traingdm | 5.4 | 11.2 | 20.5 | 12.4 |
| traingdx | 6.0 | 10.2 | 26.2 | 14.1 |
| traingda | 8.0 | 12.1 | 21.7 | 13.9 |
| trainrp | 8.9 | 12.5 | 19.6 | 13.7 |
| b. Neuron quantity in the hidden layer | ||||
| 12 | 5.0 | 10.0 | 19.4 | 11.5 |
| 14 | 4.0 | 8.2 | 25.8 | 12.7 |
| 16 | 4.0 | 13.4 | 22.5 | 13.3 |
| 18 | 13.5 | 15.0 | 28.6 | 19.0 |
| 20 | 4.0 | 8.5 | 20.1 | 10.9 |
| 22 | 4.2 | 8.9 | 24.5 | 12.5 |
| c. Quantity of hidden layers | ||||
| 3 | 5.0 | 10.0 | 19.4 | 11.5 |
| 4 | 4.5 | 8.0 | 17.8 | 10.1 |
| 5 | 3.2 | 8.1 | 11.9 | 7.7 |
| 6 | 4.6 | 7.3 | 13.9 | 8.6 |
| 7 | 5.2 | 8.8 | 17.6 | 10.5 |