Table 3.
The effect of different neural network architecture and topologies on coefficient of determination, R2, and absolute average deviation, AAD, in the estimation of lipase production obtained in the training and testing of neural networks
| Name | Model | Learning algorithm | Connection type | Transfer function output | Transfer function hidden | Training set R2 | Training set AAD (%) | Testing set R2 | Testing set AAD (%) |
| C21 | 4-16-1 | IBPa | MFFFb | Linear | Gaussian | 1 | 0.1 | 1 | 0.231 |
| D25 | 4-16-1 | IBP | MNFFc | Linear | Gaussian | 1 | 0.145 | 0.99 | 0.358 |
| C12 | 4-15-1 | IBP | MFFF | Linear | Gaussian | 1 | 0.138 | 0.953 | 0.455 |
| J22 | 4-15-1 | IBP | MNFF | Linear | Tanhd | 1 | 0.167 | 0.938 | 0.552 |
| H5 | 4-15-1 | IBP | MFFF | Linear | Tanh | 1 | 0.196 | 0.908 | 0.639 |
a Incremental Back Propagation
b Multilayer Full FeedForward
c Multilayer Normal FeedForward
d Hyperbolic Tangent Function