n coefficients estimated by the OLR model and the BLR model

y similar. The OLR model had the p value for each regression

nt, but the BLR model provided the confidence interval for each

n coefficient.

A comparison between the OLR model and the BLR model constructed for the

ntent data with ten morphological variables. The column of ߚை௅ோ represents the

coefficients estimated by the OLR model and the column of ߚ஻௅ோ represents the

coefficients estimated by the BLR model, the column of p stands for the p values

LR model, the column of lower and the column of upper stand for the lower and

dence intervals estimated at the 95% confidence level from the BLR model.

able

ߚை௅ோ

p

ߚ஻௅ோ

lower

upper

.water

−0.587834

3.22e−09

−0.59

0.09

−0.76

oleic.acid

−1.426538

0.50682

−1.34

2.16

−5.61

xide.value

−0.008017

0.94856

−0.01

0.13

−0.25

.polyphenols

0.001852

0.41356

0.00

0.00

−0.00

A.PUFA

0.024830

0.95487

0.03

0.43

−0.84

e.wieght

−1.459683

0.57864

−1.43

2.71

−6.84

e.length

−0.718050

0.12122

−0.71

0.46

−1.63

e.width

1.022776

0.45413

0.96

1.37

−1.73

weight

5.625276

0.04734

5.55

2.80

−0.05

width

0.101896

0.92096

0.14

1.03

−1.89

constrained regression analysis algorithms

d GAM do not apply a constraint to the model parameters.

e, these algorithms do not provide an efficient ranking mechanism

endent variables. In a constrained regression analysis model, the

rs of a model are assumed to follow a pre-defined distribution or

ded to a constant. Therefore the model parameters will be

ed, i.e., a competition scheme is introduced. With this

on scheme, the variables of a regression model will compete each

win within a constraint. In other words, important independent

will press down unimportant independent variables because the

r all variables is limited. When the regression coefficients of

t independent variable increase, the weights of unimportant