Page 1 of 1

Investigator ML Software for Pattern Recognition

Posted: June 11th, 2020, 12:20 pm
by sreichenbach
This past academic year, GC Image™ sponsored a student-group project to further develop Investigator ML™, a program in the GC Image software suite for multidimensional chromatography that uses machine learning (ML) to perform pattern recognition tasks such as classification, identification, regression, and other tasks. A team of six upper-division undergraduate students in the Computer Science and Engineering (CSE) Department at the University of Nebraska – Lincoln (UNL) undertook the project to extend the software’s pattern recognition capabilities from binary-class problems to multi-class problems, to add several additional pattern recognition algorithms, and to improve the software architecture and testing infrastructure. The year was especially challenging because all of their work after mid-March was done remotely, but the team did a great job and were quite successful.

Here are links to a project summary and video produced by the students:
https://cse.unl.edu/senior-design/showcase-projects


Based on that work, GC Image plans to soon release a new version of Investigator ML software to support multi-class pattern recognition from chromatographic analyses of complex sample sets. The new version:
  • Extends pattern recognition from binary-class problems to multi-class problems.
  • Implements additional ML methods.
  • Supports additional methods for data normalization, test-set generation, and cross-validation.
  • Provides additional performance metrics and new visualization tools.
Image

In addition to the ML methods in the current software — linear discriminant analysis (LDA) and k-nearest neighbors (KNN), the new version supports quadratic discriminant analysis (QDA), regularized discriminant analysis (RDA), and linear regression (LR). All of the ML methods can be applied to multi-class data sets with additional performance metrics and multi-class visualizations. Data sets can be optionally normalized by mean, range, or standard distributions; test sets can be generated randomly; and cross-validation can be performed with either leave-p-out or k-fold regimes.

If you are interested in serving as a beta-tester for the upcoming new version of GC Image Investigator ML, please send email to info@gcimage.com.