Fig. 4

NIRS ability to predict laboratory-reared mosquito species killed with Kaltox. a The ROC curve showing the false positive and true positive rates for the different classification probability thresholds, with the overall performance given by the average AUC. b Coefficient functions for each of the 100 dataset randomizations (gray lines) and the corresponding average (black line). c Histogram of the estimated linear predictor for the test mosquitoes, with the color of the bars indicating the true class, shows the model’s ability to separate the two groups of mosquitoes. The vertical black line indicates the optimum threshold for classifying mosquitoes as An. gambiae or An. coluzzii. The shaded area where the two distributions overlap corresponds to misclassified test observations, with false negatives to the left of the optimal classification threshold and false positives to the right. The confusion matrix (inset) shows the different error rates: false negative rate (fnr), false positive rate (fpr), true negative rate (tnr; An. gambiae); true positive rate (tpr; An. coluzzii). AUC, Area under the ROC curve; ROC, receiver operating characteristic