Comparative Analyses of Response Surface Methodology and Artificial Neural Network on Medium Optimization for Tetraselmis sp. FTC209 Grown under Mixotrophic Condition

Table 4

The effect of different full feed-forward network architectures on the model residual error (RMSE, 𝑅, and DC) in the prediction of lipid productivity from mixotrophic Tetraselmis sp.

Set	Model	Learning algorithm	Transfer function hidden	Transfer function output	Training set RMSE	Training set; 𝑅	Training set; DC	Testing set RMSE	Testing set; 𝑅	Testing set; DC
1	3-15-1	GA	SIGMOID	LIN	4.122	0.996	0.993	6.517	0.999	0.986
2	3-10-1	LM	TANH	LIN	4.122	0.996	0.993	8.074	0.995	0.978
3	3-15-1	LM	TANH	LIN	4.126	0.996	0.993	9.790	0.991	0.968
4	3-9-1	GA	GAUSS	LIN	4.126	0.996	0.993	10.294	0.992	0.965
5	3-10-1	GA	GAUSS	LIN	4.126	0.996	0.993	13.033	0.999	0.947