Table 2 Test results for predictive models of neonatal risk of death.

From: Development and evaluation of machine learning training strategies for neonatal mortality prediction using multicountry data

Country	Model	Algorithm	Support	Positive Outcome	CI AUC-ROC	CI Recall	Accuracy	Precision	Specificity	F1-score
General¹	General	LGBM Tuned	134,007	2,985	0.816 [0.807, 0.825]	0.220 [0.205, 0.235]	0.980	0.642	0.997	0.328
General¹	Largest train size	LGBM	134,007	2,985	0.76 [0.750, 0.771]	0.998 [0.174, 0.201]	0.980	0.657	0.188	0.292
DRC	General	LGBM Tuned	18,568	445	0.797 [0.770, 0.825]	0.245 [0.207, 0.286]	0.980	0.752	0.998	0.369
DRC	Country-specific	XGB Tuned	18,568	445	0.793 [0.765, 0.819]	0.252 [0.215, 0.293]	0.98	0.762	0.998	0.378
DRC	Largest train size	LGBM	18,568	445	0.749 [0.722, 0.778]	0.997 [0.200, 0.277]	0.979	0.679	0.238	0.353
Guatemala	General	LGBM Tuned	25,167	539	0.795 [0.772, 0.819]	0.232 [0.198, 0.269]	0.982	0.772	0.998	0.357
Guatemala	Country-specific	LGBM Tuned	25,167	539	0.796 [0.774, 0.82]	0.232 [0.197, 0.269]	0.981	0.706	0.998	0.349
Guatemala	Largest train size	LGBM	25,167	539	0.71 [0.681, 0.738]	0.999 [0.123, 0.184]	0.981	0.783	0.154	0.257
Zambia	General	LGBM Tuned	18,677	215	0.785 [0.745, 0.826]	0.265 [0.206, 0.325]	0.990	0.679	0.999	0.381
Zambia	Country-specific	LGBM Tuned	18,677	215	0.801 [0.761, 0.839]	0.237 [0.181, 0.297]	0.990	0.630	0.998	0.345
Zambia	Largest train size	LGBM	18,677	215	0.744 [0.701, 0.789]	0.998 [0.176, 0.291]	0.989	0.575	0.233	0.331
India-Belagavi	General	LGBM Tuned	16,070	313	0.784 [0.751, 0.814]	0.278 [0.224, 0.328]	0.984	0.725	0.998	0.402
India-Belagavi	Country-specific	LGBM	16,070	313	0.781 [0.75, 0.811]	0.259 [0.206, 0.308]	0.983	0.692	0.998	0.377
India-Belagavi	Largest train size	LGBM	16,070	313	0.781 [0.750, 0.811]	0.998 [0.206, 0.308]	0.983	0.692	0.259	0.377
India-Nagpur	General	LGBM Tuned	18,278	323	0.812 [0.781, 0.84]	0.232 [0.190, 0.277]	0.983	0.568	0.997	0.330
India-Nagpur	Country-specific	XGB Tuned	18,278	323	0.808 [0.778, 0.836]	0.186 [0.146, 0.229]	0.983	0.571	0.997	0.28
India-Nagpur	Largest train size	LGBM	18,278	323	0.804 [0.774, 0.833]	0.996 [0.171, 0.258]	0.982	0.504	0.214	0.300
Pakistan	General	LGBM Tuned	16,521	838	0.772 [0.752, 0.790]	0.185 [0.157, 0.212]	0.950	0.515	0.991	0.272
Pakistan	Country-specific	LGBM Tuned	16,521	838	0.770 [0.750, 0.788]	0.189 [0.161, 0.215]	0.950	0.506	0.990	0.275
Pakistan	Largest train size	LGBM	16,521	838	0.746 [0.726, 0.766]	0.997 [0.122, 0.172]	0.954	0.719	0.147	0.244
Kenya	General	LGBM Tuned	19,637	280	0.808 [0.777, 0.839]	0.161 [0.119, 0.204]	0.987	0.634	0.999	0.256
Kenya	Country-specific	XGB Tuned	19,637	280	0.805 [0.774, 0.836]	0.139 [0.102, 0.179]	0.986	0.557	0.998	0.223
Kenya	Largest train size	LGBM	19,637	280	0.764 [0.730, 0.797]	0.998 [0.109, 0.190]	0.986	0.583	0.150	0.239
Bangladesh²	General	LGBM Tuned	1,089	32	0.854 [0.779, 0.920]	0.156 [0.038, 0.306]	0.971	0.500	0.995	0.238
Bangladesh²	Largest train size	LGBM	1,089	32	0.793 [0.691, 0.896]	0.999 [0.062, 0.334]	0.975	0.857	0.188	0.308

¹The general model does not have a country-specific approach.
²Due to the lack of data on training set, Bangladesh does not present a country-specific approach, since it had no data collected in the training period.

Back to article page

Table 2 Test results for predictive models of neonatal risk of death.

Search

Quick links