Application of deep learning to predict the low serum albumin in new hemodialysis patients

Nutrition & Metabolism

Table 4 Comparison of the prediction performance of 3-months albumin average with 2 categories

Method	Model	Accuracy	Prevalence	Sensitivity	Specificity	AUC
KNN	Full	0.79	0.20	0.05	0.97	0.64
	GOA	0.79	0.20	0.11	0.96	0.61
	GOA quantile g-computation weight	0.80	0.36	0.70	0.85	0.87
SVM	Full	0.83	0.19	0.16	0.99	0.58
	GOA	0.85	0.16	0.22	0.98	0.60
	GOA quantile g-computation weight	0.88	0.37	0.82	0.91	0.86
RF	Full	0.85	0.19	0.23	0.99	0.64
	GOA	0.86	0.17	0.37	0.96	0.67
	GOA quantile g-computation weight	0.92	0.36	0.87	0.96	0.91
GBDT	Full	0.82	0.19	0.24	0.95	0.80
	GOA	0.85	0.17	0.28	0.97	0.82
	GOA quantile g-computation weight	0.88	0.36	0.78	0.94	0.95
XGBoost	Full	0.83	0.19	0.24	0.96	0.82
	GOA	0.83	0.20	0.30	0.96	0.84
	GOA quantile g-computation weight	0.88	0.35	0.79	0.93	0.94
DNN	Full	0.78	0.20	0.29	0.91	0.74
	GOA	0.79	0.20	0.24	0.93	0.73
	GOA quantile g-computation weight	0.91	0.36	0.87	0.94	0.96
Bi-LSTM	Full	0.74	0.20	0.24	0.86	0.68
	GOA	0.76	0.20	0.15	0.95	0.66
	GOA quantile g-computation weight	0.95	0.36	0.92	0.97	0.98

ISSN: 1743-7075