Faculty Mentor Information
Dr. John Kalivas (Mentor), Idaho State University
Additional Funding Sources
This material is based upon work supported by the National Science Foundation under Grant No. 2305020.
Abstract
The ability to determine whether a sample belongs to a given class or not has a large variety of applications within analytical chemistry fields. Specifically, one-class classification (OCC) is essential in many areas of study including counterfeit detection or product authentication. There are many widely accepted methods for OCC problems, such as soft independent modeling of class analogy (SIMCA) being among the most popular, all of which involve optimizing tuning parameters (number of principal components (PCs)). This poster presents an application of a new autonomous rigorous OCC named consensus OCC (Con OCC), based on a standard normal distribution of Z scores. This method uses a new similarity measure - independent of tuning parameter optimization - termed the physicochemical responsive integrated similarity measure (PRISM). PRISM combines (via data fusion) numerous similarity measures across respective tuning parameter windows into a single similarity value based on a Z distribution relative to the fused training set similarity measures. A Z distribution threshold (i.e. 2.5 standard deviations) can be used to determine whether target samples are class members or not, based on their computed PRISM scores. Presented are the classification results from a multi-instrumental beer data set representing a product authentication situation, and a chromatography olive oil data set consisting of olive oil samples and non-olive oil samples. The method of Con OCC with standard normal distribution PRISM scores provides quality classification results in terms of accuracy, sensitivity, and specificity for product authentication problems.
Authentication of Food Products Using a Standard Normal Based Consensus One-Class Classifier
The ability to determine whether a sample belongs to a given class or not has a large variety of applications within analytical chemistry fields. Specifically, one-class classification (OCC) is essential in many areas of study including counterfeit detection or product authentication. There are many widely accepted methods for OCC problems, such as soft independent modeling of class analogy (SIMCA) being among the most popular, all of which involve optimizing tuning parameters (number of principal components (PCs)). This poster presents an application of a new autonomous rigorous OCC named consensus OCC (Con OCC), based on a standard normal distribution of Z scores. This method uses a new similarity measure - independent of tuning parameter optimization - termed the physicochemical responsive integrated similarity measure (PRISM). PRISM combines (via data fusion) numerous similarity measures across respective tuning parameter windows into a single similarity value based on a Z distribution relative to the fused training set similarity measures. A Z distribution threshold (i.e. 2.5 standard deviations) can be used to determine whether target samples are class members or not, based on their computed PRISM scores. Presented are the classification results from a multi-instrumental beer data set representing a product authentication situation, and a chromatography olive oil data set consisting of olive oil samples and non-olive oil samples. The method of Con OCC with standard normal distribution PRISM scores provides quality classification results in terms of accuracy, sensitivity, and specificity for product authentication problems.