Automated early detection of drops in commercial egg production using neural networks

Figures & data

Figure 1. Multilayer perceptron with one hidden layer representation.

Figure 2. Egg production boxplot representing the daily average per bird for each of the 24 flocks analysed in this work.

Figure 3. Daily production per bird in representative flocks number 8 and 23.

Figure 4. Grid search for window size and feature selection threshold: (a) accuracy, (b) specificity, (c) sensitivity and (d) positive predictive value.

Figure 5. Selected features with a feature selection threshold of 65 in a window of size equal to 18 d.

Table 1. Multiple comparisons of the evaluated architectures.

Network	Accuracy		Specificity		Sensitivity		Positive predictive value
architecture	Mean	STD	Mean	STD	Mean	STD	Mean	STD
[no HL]	0.955^a	0.062	0.956^a	0.063	0.900^a	0.083	0.387^b	0.159
[25]	0.968^a	0.021	0.970^a	0.022	0.880^a	0.057	0.451^a	0.174
[25,25]	0.961^a	0.026	0.962^a	0.027	0.884^a	0.062	0.411^ab	0.178
[25,50]	0.959^a	0.028	0.961^a	0.029	0.884^a	0.058	0.405^ab	0.175
[50]	0.967^a	0.020	0.970^a	0.021	0.879^a	0.059	0.450^a	0.175
[50,25]	0.962^a	0.026	0.964^a	0.027	0.876^a	0.070	0.418^ab	0.180
[50,50]	0.962^a	0.025	0.964^a	0.026	0.876^a.	0.061	0.416^ab	0.175

Means in a row with no common superscript letter differ significantly according to Tukey’s Honest Significance Test for a value of P < 0.01.

Figure 6. Performance chart of the model for different values of the parameter S.

Table 2. Performance metrics for 5 prediction intervals.

	Prediction interval
Performance	0 d	1 d	2 d	3 d	4 d	5 d
Accuracy	0.9925 ± 0.003	0.9896 ± 0.005	0.9864 ± 0.006	0.983 ± 0.009	0.9797 ± 0.01	0.9758 ± 0.012
Specificity	0.9954 ± 0.003	0.9936 ± 0.005	0.9923 ± 0.006	0.9905 ± 0.009	0.9891 ± 0.011	0.9874 ± 0.014
Sensitivity	0.8639 ± 0.061	0.8169 ± 0.078	0.7362 ± 0.091	0.6643 ± 0.104	0.5795 ± 0.132	0.4821 ± 0.152
Positive predictive value	0.8125 ± 0.119	0.7564 ± 0.14	0.7134 ± 0.166	0.6723 ± 0.19	0.6246 ± 0.22	0.567 ± 0.233

Figure 7. Boxplots of performance metrics for prediction intervals: (a) accuracy, (b) specificity, (c) sensitivity and (d) positive predictive value.