Table 2.

Clustering performance on interval evaluation indexes based on various patient representations.

Representation schemes	Parameters for training				Cluster evaluation indexes
	Corpus used	Corpus with shuffling	Window size	Hopkins statistic		Silhouette index	Davies-Bouldin index
Embedding-based representation	Full	Yes	5	0.922		0.783	1.067
	Stroke	Yes	5	0.913		0.862a	0.551b
	Full	No	5	0.903		0.685	1.711
	Stroke	No	5	0.925		0.672	1.382
	Full	No	255	0.922		0.783	1.065
	Stroke	No	224	0.931c		0.790	0.772
Multi-hot representationd	N/Ae	N/A	N/A	0.813		0.233	3.236
Mixture representationf	N/A	N/A	N/A	0.918		0.141	4.157

^aHighest value of the Silhouette index.

^bLowest value of the Davies-Bouldin index.

^cHighest value of the Hopkins statistic.

^dMulti-hot representation: representation method of the combinations of one-hot codes.

^eN/A: not applicable.

^fMixture representation: representation method of the combination of multi-hot codes for discrete features and real numbers for continuous values of age and laboratory tests.