SIMCA is a supervised chemometric method based on principal
component analysis (PCA). It requires assigning training data sets to
classes. Thirty cornmeal samples were used to train the classification
models while eight samples which were randomly selected
were used as an independent validation set (five conventional and
three organic cornmeal). Samples were divided into calibration and
independent validation sets in away that all replicates belonging to
the same cornmeal sample were either used only in the calibration
or validation set. In this study, 2 classes were designed: conventional
cornmeal (class 1) and organic cornmeal (class 2). The
number of PCs used in the model was determined by a crossvalidation
procedure (leave-one-out) and confidence intervals or
class boundaries were calculated by means of residual standard
deviations to assess class membership and detecting outliers.
Probability clouds (a ¼ 0.05) surrounding the clusters allow SIMCA
to be used as a predictive modeling system. In addition, residuals
provide valuable information regarding class homogeneity, separation
between classes (interclass distance), and the relative
strength of any given variable to model the structure of a class or to
discriminate between classes (discriminating power). SIMCA's
interclass distance (ICD) describes quantitatively the similarity or
dissimilarity of the different classes, being generally accepted that
samples can be differentiated when ICD > 3 (Vogt and Knutsen,
1985). The discriminating power of variables may be used to
eliminate noise from the data set, such that variables having both
low discriminating power and modeling power can be eliminated.
SIMCA performance was examined in terms of discriminating power,
class projections, misclassifications (percentage of samples
correctly allocated to their original groups) and ICD.
SIMCA is a supervised chemometric method based on principal
component analysis (PCA). It requires assigning training data sets to
classes. Thirty cornmeal samples were used to train the classification
models while eight samples which were randomly selected
were used as an independent validation set (five conventional and
three organic cornmeal). Samples were divided into calibration and
independent validation sets in away that all replicates belonging to
the same cornmeal sample were either used only in the calibration
or validation set. In this study, 2 classes were designed: conventional
cornmeal (class 1) and organic cornmeal (class 2). The
number of PCs used in the model was determined by a crossvalidation
procedure (leave-one-out) and confidence intervals or
class boundaries were calculated by means of residual standard
deviations to assess class membership and detecting outliers.
Probability clouds (a ¼ 0.05) surrounding the clusters allow SIMCA
to be used as a predictive modeling system. In addition, residuals
provide valuable information regarding class homogeneity, separation
between classes (interclass distance), and the relative
strength of any given variable to model the structure of a class or to
discriminate between classes (discriminating power). SIMCA's
interclass distance (ICD) describes quantitatively the similarity or
dissimilarity of the different classes, being generally accepted that
samples can be differentiated when ICD > 3 (Vogt and Knutsen,
1985). The discriminating power of variables may be used to
eliminate noise from the data set, such that variables having both
low discriminating power and modeling power can be eliminated.
SIMCA performance was examined in terms of discriminating power,
class projections, misclassifications (percentage of samples
correctly allocated to their original groups) and ICD.
การแปล กรุณารอสักครู่..