Fig. 21

Scatter plots illustrating the performance of three NN configurations with differing layer distributions but an equal total number of hidden neurons, expressed as powers of two. The left panel shows MAE vs. total loss, and the right shows MSE vs. total loss. The shaded bands around each trend line indicate the inter-quartile range, reflecting variability across runs. Blue-highlighted architectures demonstrate more pronounced reductions in MAE and MSE for comparable decreases in total loss, emphasizing the critical role of neuron allocation in determining model accuracy.