Meaning
Chemometric classification algorithms construct individual principal component models for each sample category in a multidimensional dataset. Analytical chemists apply the soft independent modeling class analogy to classify incoming resin lots based on their near-infrared spectra. This mathematical classifier operates by evaluating each class independently, allowing a sample to be assigned to multiple classes or to none.
Model Construction
Training sets of known conforming and non-conforming resin batches are compiled to build the individual class models. In the soft independent modeling class analogy, the number of principal components retained for each class is optimized to capture the specific variance of that category. This localized model building ensures that the classifier can identify subtle differences between similar polymer grades.
Classification Threshold
Acceptance boundaries are established for each class using a combination of the F-test and the mahalanobis distance in the score space. The soft independent modeling class analogy calculates the distance of a new sample from each class model to determine its classification. If a sample falls within the acceptance region of a class, it is classified as belonging to that group, while samples that fall outside all regions are flagged as unknown.
This capability prevents the misclassification of novel contaminants as known materials.
Process Integration
Industrial spectrometers use this algorithm to automatically screen raw materials as they are received at the factory dock. The soft independent modeling class analogy provides a rapid, automated pass or fail decision that is integrated with the warehouse management software. This integration prevents the incorrect routing of material and reduces the risk of contamination in the production process.
