Table 1 Performance of models for predicting stillbirth using different classification algorithms and 10-fold cross validation.

From: Stillbirth risk prediction using machine learning for a large cohort of births from Western Australia, 1980–2015

Classifiers	Model	AUC	5% FPR						10% FPR
Classifiers	Model	AUC	+LR	−LR	Sensitivity	PPV	NPV	CorrectlyClassified	+LR	−LR	Sensitivity	PPV	NPV	Correctly Classified
Logistic Regression	A	0.830	8.10	0.63	40.5	4.72	99.62	94.67	5.52	0.50	55.2	3.26	99.7	89.79
	B	0.834	8.07	0.63	40.5	4.32	99.65	94.68	5.57	0.49	55.7	3.02	99.73	89.80
	C	0.811	7.59	0.66	37.8	3.89	99.65	94.72	5.14	0.54	51.6	2.67	99.71	89.75
	D	0.602	2.25	0.93	11.2	1.35	99.43	94.49	1.90	0.90	19.0	1.15	99.45	89.57
	E	0.633	3.29	0.88	16.5	1.80	99.51	94.54	2.44	0.84	24.4	1.35	99.53	89.64
	F	0.799	6.02	0.74	30.1	3.26	99.59	94.64	4.65	0.60	46.4	2.53	99.67	89.76
Decision Tree	A	0.819	8.16	0.62	40.7	4.75	99.62	94.67	5.68	0.51	54.1	3.35	99.69	90.24
	B	0.808	8.18	0.63	40.6	4.38	99.65	94.73	5.01	0.51	54.7	2.73	99.72	88.88
	C	0.776	6.98	0.68	35.8	3.59	99.64	94.58	5.19	0.63	42.3	2.69	99.67	91.40
	D	0.589	2.07	0.95	10.2	1.25	99.43	94.54	1.78	0.91	17.7	1.08	99.45	89.60
	E	0.599	3.16	0.89	15.2	1.73	99.50	94.68	2.33	0.86	23.0	1.29	99.52	89.67
	F	0.779	5.94	0.74	30.1	3.22	99.59	94.58	5.71	0.73	31.2	3.09	99.59	94.13
Random Forest	A	0.831	8.12	0.63	40.6	4.73	99.62	94.67	5.55	0.50	55.5	3.28	99.70	89.79
	B	0.836	8.22	0.62	41.1	4.40	99.65	94.71	5.66	0.48	56.4	3.07	99.73	89.85
	C	0.788	7.29	0.67	36.4	3.74	99.64	94.69	4.91	0.57	49.1	2.55	99.70	89.78
	D	0.594	2.09	0.94	10.4	1.26	99.43	94.48	1.75	0.92	17.5	1.06	99.44	89.57
	E	0.633	2.87	0.90	14.4	1.58	99.50	94.54	2.37	0.85	23.7	1.31	99.53	89.64
	F	0.801	5.96	0.74	29.8	3.23	99.59	94.64	4.66	0.59	46.7	2.54	99.67	89.76
XGBoost	A	0.840	8.93	0.58	44.6	5.18	99.65	94.70	5.81	0.47	58.1	3.43	99.72	89.81
	B	0.842	9.03	0.58	45.3	4.81	99.68	94.71	5.86	0.46	58.7	3.18	99.74	89.82
	C	0.804	7.54	0.66	37.6	3.86	99.65	94.69	5.12	0.54	51.2	2.66	99.71	89.81
	D	0.596	2.18	0.94	10.9	1.32	99.43	94.49	1.85	0.91	18.5	1.12	99.45	89.57
	E	0.628	3.31	0.88	16.6	1.82	99.51	94.55	2.47	0.84	24.7	1.36	99.53	89.64
	F	0.805	6.56	0.71	32.8	3.54	99.61	94.66	4.84	0.57	48.4	2.64	99.68	89.76
Multi-layer Perceptron	A	0.836	8.57	0.60	42.8	4.98	99.63	94.69	5.65	0.48	56.5	3.34	99.71	89.80
	B	0.840	8.69	0.60	43.5	4.64	99.67	94.71	5.73	0.48	57.2	3.11	99.73	89.83
	C	0.801	7.38	0.67	36.7	3.78	99.65	94.72	5.12	0.55	50.9	2.65	99.71	89.84
	D	0.595	2.15	0.94	10.8	1.30	99.43	94.49	1.84	0.91	18.4	1.11	99.45	89.56
	E	0.634	3.24	0.88	16.2	1.78	99.51	94.57	2.41	0.84	24.1	1.33	99.53	89.64
	F	0.802	6.43	0.71	32.1	3.47	99.60	94.65	4.81	0.58	48.1	2.62	99.68	89.77

Estimates with 95% confidence intervals are provided in the Supplementary Material, Supplementary Table 6.
Model A – Socio-demographics, chronic conditions, current pregnancy complications and characteristics.
Model B – Predictors from Model A, plus previous pregnancy history.
Model C – Predictors from Model A, plus grandmother’s pregnancy history, parental birth outcomes.
Model D – Predictors known at the booking appointment.
Model E – Predictors from Model D, plus previous pregnancy history.
Model F – Predictors from Model E, plus current pregnancy complications and characteristics.
Abbreviations: AUC – Area under the receiving-operator characteristic curve; +LR – Positive likelihood ratio; -LR – Negative likelihood ratio; FPR – alpha (type I error) = 1-specificity; Sensitivity – detection rate, TPR; TP – True Positives; FP – False Positives; TN – True Negatives; FN – False Negatives; PPV – Positive predictive value = TP/(TP + FP); NPV - Negative predictive value = TN/(FN + TN); CI – Confidence Interval.

Back to article page

Table 1 Performance of models for predicting stillbirth using different classification algorithms and 10-fold cross validation.

Search

Quick links