fig5
Figure 5. Experimental results of hyperparameter sensitivity. (A-E) MSE variations across key hyperparameters: GRU hidden dimensions, GRU layers, embedding dimensions, MHA heads, and FFN hidden dimensions; (F) Normalized importance scores of investigated hyperparameters. MSE: Mean squared error; GRU: gated recurrent unit; MHA: multi-head attention; FFN: feed-forward network.
