CPACT Webinar on The Intersection of Machine Learning, Chemometrics, and Spectroscopy

Presented by Brian Rohrback of Infometrix, Inc.

April 23, 2026 (7:00PM UK Time).

See abstract below. Visit CPACT Webinars or contact info@infometrix.com for details

AI and machine learning have stormed into our scientific and marketing lexicons.  As we discuss the integration into analytical chemistry applications, we face the invariable need to merge with the field of chemometrics.  We know chemometrics as an area of study that has generated a set of tools for practitioners to use in extracting the information content from sets of analytical data.  Machine learning is the extension of this idea, just without human intervention.  As we employ the tools provided by chemometrics to autonomously automate a process, where the computer is making decisions based on the input data, the chemometrics becomes a cog in the machine learning world.  One area ripe for this combination is optical spectroscopy, particularly IR, NIR, and Raman.

Let’s do a thought experiment.  What if we decided we wanted to fully automate the use of optical spectroscopy for a quality control application?  What would be required to take any spectroscopy instrument, put it into a lab or process stream, have it learn the application, build an optimized model, deploy the model for QC, and maintain the calibration for the life of the instrument. Can this be done without human interaction?

To standardize the control of spectroscopy assessments, there are four primary software-related areas to tackle, two of which the user may only need to do once.

1. At the start, a method needs to be set that optimizes how future spectra will be manipulated and involves algorithm selection, choice of preprocessing, and potentially trimming the wavelength range.
2. The other early process is to understand the precision of the laboratory methods, how they impact calibration models, and how this information needs to be factored into understanding system performance.
3. On a continuous basis, the process chemistry can change dictating a maintenance effort to determine the optimum number of factors and identify outliers that negatively impact model performance.
4. A system has been outlined by ASTM to automatically flag when the model performance has degraded.

Infometrix has spent the last decade and a half commercializing a system designed to fully automate and efficiently optimize all aspects of the above calibration. Components of the thought experiment are in place and a discussion of the approach (plus solutions to encounters with implementation quicksand) shows how to blend chemometrics into machine learning for the benefit of industry.