Figure 3.
Panel a contains an input layer with Bias, C S R at N R I 50.8 per cent, G I C at N R I 74.7 per cent, G T L at N R I 87.6 per cent, and S P at N R I 100 per cent. The sigmoid hidden layer contains Bias, H 1 1, H 1 2, and H 1 3. Bias has negative weights to all 3 hidden nodes. C S R, G I C, and S P have positive weights to the 3 hidden nodes. G T L has positive weights to H 1 1 and H 1 2, and a negative weight to H 1 3. Hidden Bias, H 1 1, and H 1 3 have positive weights to G H R M, while H 1 2 has a negative weight. The output signal applies sigmoid activation. Panel b contains Bias and G H R M as inputs, with G H R M carrying beta 0.570. The sigmoid hidden layer contains Bias, H 1 1, and H 1 2. Bias has negative weights to both hidden nodes, while G H R M has positive weights to both. Hidden Bias, H 1 1, and H 1 2 have positive weights to C S P. The output signal applies sigmoid activation. Solid connections denote positive weights, and broken connections denote negative weights.(a) ANN Model 1 – Architecture for Predicting GHRM (b) ANN Model 2 – Architecture for Predicting CSP from GHRM