Procedure code overutilization detection from healthcare claims using unsupervised deep learning methods

Table 2 Performance metrics of models trained on 100k_claims dataset

Test Datasets	Metrics	DBSCAN	Autoencoder
Out-of-sample	Precision	0.22	0.87
	Recall	0.58	1.0
	F1	0.32	0.93
Manually annotated	Precision	0.19	0.36
	Recall	0.41	0.86
	F1	0.26	0.51

ISSN: 1472-6947