. 2019 Dec 17;7(4):e14782. doi: 10.2196/14782

Table 2.

Comparisons of the performance of reusing models with different semantic similarity levels. Similarity threshold: 0.01; DBScan EPS: 0.38. Reusing models trained for more (semantically) similar phenotypes achieved adaptation results with less effort (more duplicate waste identified) in all cases, and the results were more accurate in three of four cases. Performance metrics of better reusable models are highlighted as bold numbers.

Model reuse cases	Duplicate waste	Macro-accuracy	Micro-accuracy
Diabetes by Type 2 Diabetes^a	0.502^b	0.966^b	0.933^b
Diabetes by Hypercholesterolemia	0.477	0.965	0.930
Stroke by Heart Attack^a	0.711^b	0.948^b	0.955^b
Stroke by Fatigue	0.220	0.884	0.938
Heart attack by Infarct^a	0.569^b	0.989^b	0.966^b
Heart attack by Bruise	0.529	0.821	0.889
Multiple Sclerosis by Myasthenia Gravis^a	0.761^b	0.944	0.971
Multiple Sclerosis by Diabetes	0.522	0.993^b	0.979^b

^aMore similar model reuse cases.

^bPerformance metrics of better reusable models.