Fig 1.
Average spectra of laboratory sterile-unmarked flies (n = 114) and wild-caught flies (n = 143), each being sampled at discrete wavelengths in the interval [500, 2400 nm].
Fig 2.
NIRS ability to discriminate between sterile unmarked and wild-caught flies.
(a) Receiver operating characteristic (ROC) curve illustrating the diagnostic ability of the best-fit model. Overall performance is given by the average area under the ROC curve (AUC). A theoretical perfect diagnostic would be in the top left corner. The average ROC curve shown by the solid line with boxplots shows the variability for 100 randomizations of the training, validation and testing. (b) Coefficient functions for the best-fit model for each of the 100-dataset randomizations (grey lines) and the overall average (black line). (c) Histogram of the estimated linear predictor for the test observations, color-coded by the true class: marked irradiated flies (light blue colored bars) or Wild-caught flies (green bars). Vertical solid black line indicates the best threshold for differentiating marked irradiated flies and wild-caught flies. Darker blue bars indicate where the two distributions overlap and misclassified flies’ status. Misclassified wild-caught Glossina are shown to the left of the optimal classification threshold line and misclassified sterile unmarked Glossina to the right. Inset shows the confusion matrix illustrating the different error rates: true negative rate (tnr) for the sterile unmarked flies correctly classified; false negative rate (fnr) for the misclassified field flies; false positive rate (fpr) for the misclassified sterile unmarked flies; and true positive rate (tpr) corresponding to the field flies correctly classified.
Fig 3.
NIRS accuracy in predicting laboratory-reared and wild-caught Glossina.
a: Marking effect (sterile-unmarked vs sterile-marked); b: Irradiation effect (fertile unmarked vs sterile unmarked); c: effect (sterile marked vs wild); d: Origin effect (laboratory fertile and unmarked vs wild).