Analysis of near infrared spectra for age-grading of wild populations of Anopheles gambiae

Parasites & Vectors

Table 2 Calibration, cross-validation, validation and independent test set 1 (ITS1) results for each algorithm on the 6 datasets

Dataset	Samples	No. var	RMSEC	R²Cal	RMSECV	R²CV	LV	RMSEV	RMSEP-ITS1
Dataset 1
PLS	178	1851	2.68	0.55	3.16	0.39	10	2.90	3.88
iPLS	178	180	2.41	0.64	2.92	0.55	10	2.97	5.52
enPLS	175	400	1.71	0.82	2.04	0.74	na	2.62	7.01
MASS	173	258	2.00	0.74	2.28	0.66	10	2.93	4.04
VCPA	178	11	2.36	0.65	2.52	0.60	10	3.11	4.64
svmLinear	178	1851	na	na	2.83	0.59	na	2.70	4.29
Dataset 2
PLS	156	1851	1.85	0.83	2.28	0.74	10	2.71	4.08
iPLS	156	120	1.54	0.93	1.20	0.90	10	2.41	3.88
enPLS	152	300	0.81	0.97	1.05	0.95	na	1.89	4.19
MASS	153	385	0.87	0.96	1.10	0.94	10	2.41	4.33
VCPA	156	10	1.88	0.82	2.08	0.78	10	2.49	3.29
svmLinear	156	1851	na	na	1.89	0.81	na	2.13	4.60
Dataset 3
PLS	160	1851	2.05	0.80	2.61	0.70	10	2.85	5.53
iPLS	160	60	1.97	0.81	2.41	0.78	10	2.29	5.61
enPLS	158	350	0.76	0.97	1.44	0.90	na	1.96	4.29
MASS	158	441	1.24	0.93	1.59	0.88	10	2.06	4.17
VCPA	160	10	1.95	0.82	2.05	0.80	8	2.55	3.40
svmLinear	160	1851	na	na	1.94	0.82	na	2.23	3.76
Dataset 4
PLS	200	1851	2.10	0.76	2.60	0.64	10	2.43	5.18
iPLS	200	60	1.71	0.84	2.17	0.80	10	2.41	4.05
enPLS	195	350	0.85	0.96	1.32	0.90	na	1.49	3.56
MASS	196	140	1.55	0.87	1.78	0.82	10	1.98	3.95
VCPA	200	11	2.28	0.71	2.39	0.69	7	2.72	6.44
svmLinear	200	1851	na	na	1.99	0.77	na	1.74	4.32
Dataset 5
PLS	334	1851	2.94	0.50	3.16	0.43	10	3.42	3.57
iPLS	334	180	2.50	0.64	2.76	0.58	10	2.72	6.70
enPLS	330	200	1.77	0.82	2.07	0.75	na	3.10	4.69
MASS	329	466	2.20	0.71	2.36	0.67	10	3.10	3.67
VCPA	334	12	2.82	0.54	2.89	0.51	8	3.70	4.79
svmLinear	334	1851	na	na	2.66	0.63	na	2.81	3.70
Dataset 6
PLS	494	1851	3.24	0.43	3.50	0.34	10	3.29	3.43
iPLS	494	120	3.21	0.44	3.36	0.41	8	2.99	5.01
enPLS	479	300	1.76	0.83	2.21	0.73	na	2.77	3.33
MASS	492	482	2.58	0.64	2.83	0.56	10	3.08	2.96
VCPA	494	10	3.43	0.47	3.15	0.46	10	3.43	2.48
svmLinear	494	1851	na	na	2.68	0.61	na	2.78	3.49

Note: eEach of the six datasets were used to generate models using six regression algorithms. The root mean squared error (RMSE) is presented for the calibration, cross-validation and validation sets, and independent test set 1. This measure (with units of “days”) allows for an approximation of how much error is present across the range of ages present in each dataset
Abbreviations: No. of var. number of variables used, RMSEC root mean squared error of calibration, R ² Cal coefficient of variation of calibration, RMSECV root mean squared error of cross-validation, R ² CV coefficient of variation of cross-validation based on the actual vs predicted ages of the average of the 5 or 10 fold cross-validation, LV number of latent variables used in PLS regression (if applicable), RMSEV root mean squared error of validation set, RMSEP-ITS1 root mean squared error of prediction for independent test set 1, na not available for RMSEC/ R²Cal values (was not calculated natively in the implementation of svmLinear) or not applicable for LV (due to use of ensemble models in enPLS and not used in support vector machines)

ISSN: 1756-3305