Table 3.
Performance metrics of Artificial Neural Networks classifiers.
| Pre-processing | Model | Seven-Class Models |
Three-Class Models |
Two-Class/Binary Models |
|||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Optimal Parameters | ACC.cv | ACC.p | Optimal Parameters | ACC.cv | ACC.p | Optimal Parameters | ACC.cv | Sens. cv | Prec. cv | Spec. cv | F1.cv | ACC.p | Sens. p | Prec. p | Spec. p | F1.p | MCC.p | ||
| Unprocessed | ANN | dec = 0.001,size = 2 | 73.6 | 78.9 | dec = 0.001,size = 4 | 100 | 98.9 | dec = 0.001,size = 2 | 99.7 | 100 | 99.7 | 94.5 | 99.8 | 99.7 | 99.7 | 100 | 100 | 99.8 | 0.98 |
| SG smoothing | ANN | dec = 0.001,size = 3 | 83.6 | 79.8 | dec = 0.001,size = 4 | 100 | 100 | dec = 0.001,size = 2 | 99.6 | 100 | 99.6 | 92.5 | 99.8 | 99.7 | 99.7 | 100 | 100 | 99.8 | 0.98 |
| SG+1st deriv. | ANN | dec = 0.001,size = 3 | 87.7 | 76.6 | dec = 0.001,size = 3 | 99.9 | 100 | dec = 0.001,size = 1 | 99.8 | 99.9 | 99.9 | 98.0 | 99.9 | 99.8 | 99.8 | 100 | 100 | 99.9 | 0.99 |
| SG+2nd deriv. | ANN | dec = 0.001,size = 3 | 87.9 | 78.3 | dec = 0.001,size = 2 | 99.6 | 100 | dec = 0.001,size = 1 | 99.8 | 99.9 | 99.9 | 99.0 | 99.9 | 99.8 | 99.8 | 100 | 100 | 99.9 | 0.99 |
| SNV | ANN | dec = 0.01,size = 3 | 92.0 | 79.5 | dec = 0.01,size = 3 | 100 | 100 | dec = 0.01,size = 1 | 99.9 | 99.9 | 99.9 | 98.5 | 99.9 | 99.8 | 99.8 | 100 | 100 | 99.9 | 0.99 |
| SNV + SG Smoothing | ANN | dec = 0.001,size = 2 | 73.0 | 68.1 | dec = 0.01,size = 3 | 100 | 100 | dec = 0.01,size = 1 | 99.9 | 99.9 | 99.9 | 98.5 | 99.9 | 99.8 | 99.8 | 100 | 100 | 99.9 | 0.99 |
| SNV + SG+1st deriv. | ANN | dec = 0.001,size = 3 | 93.7 | 83.4 | dec = 0.001,size = 2 | 99.8 | 100 | dec = 0.001,size = 1 | 99.9 | 100 | 99.9 | 98.5 | 99.9 | 100 | 100 | 100 | 100 | 100 | 1.00 |
| SNV + SG+2nd deriv. | ANN | dec = 0.001,size = 3 | 93.2 | 77.9 | dec = 0.001,size = 2 | 99.9 | 100 | dec = 0.001,size = 1 | 99.9 | 100 | 99.9 | 97.0 | 99.9 | 100 | 100 | 100 | 100 | 100 | 1.00 |
| MSC | ANN | dec = 0.01,size = 2 | 71.6 | 57.1 | dec = 0.001,size = 1 | 99.2 | 98.7 | dec = 0.001,size = 2 | 99.7 | 99.9 | 99.7 | 93.5 | 99.8 | 99.8 | 99.8 | 100 | 100 | 99.9 | 0.99 |
| MSC + SG Smoothing | ANN | dec = 0.001,size = 3 | 84.8 | 81.1 | dec = 0.001,size = 1 | 99.2 | 99.3 | dec = 0.001,size = 2 | 99.6 | 99.9 | 99.6 | 91.5 | 99.8 | 99.7 | 99.7 | 100 | 100 | 99.8 | 0.98 |
| MSC + SG+1st deriv. | ANN | dec = 0.001,size = 3 | 87.7 | 76.6 | dec = 0.001,size = 3 | 99.9 | 100 | dec = 0.001,size = 1 | 99.8 | 99.9 | 99.9 | 98.0 | 99.9 | 99.8 | 99.8 | 100 | 100 | 99.9 | 0.99 |
| MSC + SG+2nd deriv. | ANN | dec = 0.001,size = 3 | 83.9 | 77.4 | dec = 0.001,size = 3 | 99.7 | 100 | dec = 0.001,size = 1 | 99.8 | 99.9 | 99.9 | 97.2 | 99.9 | 100 | 100 | 100 | 100 | 100 | 1.00 |
The metric values for the trained models represent averaged classification parameters of 10-fold cross-validation repeated ten times. ACC.cv = Accuracy, Sens.cv = Sensitivity, Prec.cv = Precision, Spec.cv = Specificity, and F1.cv = F1 Score for cross-validation. ACC.p = Accuracy, Sens.p = Sensitivity, Prec.p = Precision, Spec.p = Specificity, and F1.p = F1 Score for the external validation set (test set). SNV = Standard Normal Variate; MSC = Multiplicative Scatter Correction; SG = Savitzky-Golay smoothing; 1st deriv. = 1st derivative; 2nd deriv. = second derivative. Size is the number of optimal number of neurons in the hidden layers selected based on cross-validation and oneSE rule. Dec = decay, regularization parameter. For the Seven-Class system, the classification involves seven groups: extra-virgin olive oil (EVOO), hazelnut oil (HZO), olive pomace oil (POO), refined olive oil (ROO), EVOO + HZO, EVOO + POO, and EVOO + ROO. The Three-Class system categorizes oils into three groups: authentic extra-virgin olive oil, edible oil adulterant (100%), or adulterated (1–40% adulteration) olive oil. The Two-Class system is a binary classification distinguishing between pure EVOO and adulterated olive oil (1–100% adulteration).