Table 2 Predictive performance of chemometric models with variable selection algorithms for peroxide and iodine value prediction in crude palm oil

From: Machine learning-assisted Raman spectroscopy for non-destructive analysis of crude palm oil quality

Parameter

Models

PCs

Number of variables

Calibration set n = 133

Prediction set n = 67

RPD

Rc

RMSEC

Rp

RMSEP

Peroxide value (meq O2/kg)

PLS

6

1024

0.9597

0.5154

0.9481

0.6367

4.8935

CARS-PLS

9

104

0.9794

0.4701

0.9557

0.5649

2.3079

UVE-PLS

6

66

0.9705

0.5104

0.9613

0.5186

3.6138

GA-PLS

6

148

0.9641

0.5603

0.9471

0.6824

3.0003

Iodine value (g I2/100 g)

PLS

6

1024

0.9692

0.3565

0.9400

0.5002

3.9139

CARS-PLS

8

91

0.9687

0.3161

0.9483

0.5340

2.3208

UVE-PLS

7

45

0.9724

0.5266

0.9650

0.6203

3.7535

GA-PLS

6

140

0.9647

0.5945

0.9327

0.8678

2.7623

  1. n number of samples, PCs principal components, PLS partial least squares, RPD residual predicted deviation, Rc correlation coefficient for the calibration set, RMSECV root mean square error of cross-validation, Rp correlation coefficient for the prediction set, RMSEP root mean square error of prediction, UVE-PLS uninformative variable elimination partial least squares, CARS-PLS competitive adaptive reweighted sampling partial least squares, GA-PLS genetic algorithm partial least squares.
Back to article page