Table 7.

Evaluation metrics for various representative hyperparameter configurations.

Layers	Activation	Pearson	Spearman	Procrustes	GMM	Comments
-	-	0.916	0.896	0.272	0.814	PCA
[ ]	ReLU	0.940	0.920	0.226	0.695	The approximate linearity of the ReLU activation function of this model favours distance preservation
[ ]	Sigmoid	0.917	0.906	0.240	0.543	The non-linearity of the Sigmoid activation affects distance metrics slightly and improves density metrics
[3]	Sigmoid	0.840	0.830	0.301	0.321	It balances distance preservation and density metric results
[5,4,3]	ReLU	0.635	0.622	0.505	0.104	It is a complex model with good density metric results but produces dense points in the latent dimension not apt for visualisation of patient trajectories over time. In addition, distance metric results show that distances are not preserved and therefore it is inadequate for similarity-based retrieval