Table 7.

Predictive performance of the selected binary logistic regression model for detecting woody breast (WB) condition in broiler carcasses by strain at different live weights using image measurements.

Strain	Live weight (kg)	Data set1	Fit details2							Parameter estimates and odds ratio (OR)
			Gen. R2	RMSE	MAD	MR (%)	TPR (%)	FPR (%)	AUC	Parameter	Estimate	Std. Error	Pr > χ2	OR
High breast-yielding (HBY)	<3.402	Training	0.62	0.19	0.07	4.43	61.54	1.38	0.96	M1	2.27	1.03	0.0265	9.72
		Validation	0.74	0.16	0.06	4.76	60.00	1.72	0.99	M2	−9.37	3.38	0.0055	8.49 × 10−5
										M3	2.11	0.80	0.0086	8.26
	≥3.402	Training	0.62	0.31	0.19	12.41	71.62	6.25	0.93	M1	1.56	0.38	<0.0001	4.75
		Validation	0.61	0.32	0.19	11.50	77.42	7.32	0.92	M2	−8.92	1.52	<0.0001	1.34 × 10−4
										M3	1.75	0.35	<0.0001	5.74
Standard breast-yielding (SBY)	≥3.402	Training	0.57	0.24	0.11	8.09	56.25	3.33	0.96	M1	1.49	0.66	0.0243	4.43
		Validation	0.66	0.22	0.09	7.94	57.14	3.57	0.97	M2	−7.04	2.39	0.0032	8.77 × 10−4
										M3	1.54	0.53	0.0038	4.65

Abbreviations: AUC, area under the ROC (receiver operating characteristic) curve; FPR, false positive rate; MAD, mean absolute deviation; MR, misclassification rate; RMSE, root mean square error; TPR, true positive rate (sensitivity, ST).

¹Training (HBY, < 3.402 kg: n = 158; HBY, ≥ 3.402 kg: n = 266; SBY, ≥ 3.402 kg: n = 136) and validation (HBY, < 3.402 kg: n = 63; HBY, ≥ 3.402 kg: n = 113; SBY, ≥ 3.402 kg: n = 63).

²The model used for WB prediction (model 1): Logit (p) = α + β1 M1 + β2 M2 + β3 M3.