Enhancing microplastic identification using machine learning and Raman Spectroscopy

By Úna Hogan 

Researchers in the Smith Group at the University of Waterloo, in collaboration with the Microplastics Fingerprinting project, have developed a new machine learning model to improve microplastic identification. This model is based on a k-Nearest Neighbors (kNN) approach and analyzes a library of Raman spectra from various plastics to enhance accuracy.  

How Raman Spectroscopy Identifies Plastics 

Raman spectroscopy identifies chemical compounds by analyzing their unique spectral fingerprints. However, environmental degradation can alter these fingerprints, making plastic identification more difficult. UV exposure, thermal degradation, chemical reactions (e.g., with pesticides or cleaners), accumulation of dirt, and physical weathering all contribute to spectral changes that hinder matching samples using traditional libraries.  

Machine learning offers a more effective solution.  

A Smarter Approach to Plastic Classification 

Previous machine learning models focused only on the lower region of the Raman spectrum which often failed to differentiate plastics with similar chemical structures. The new kNN model analyzes both the lower and the higher spectral region, improving accuracy and reducing misidentifications.  

The model achieved an F1 score of 0.97, demonstrating near-perfect classification performance.  

Advancing Microplastic Research 

By improving identification accuracy, this research helps scientists better track and understand microplastic pollution. As machine learning continues to refine analytical techniques, researchers can more effectively assess environmental contamination and develop stronger strategies for reducing plastic waste. 

This research was made possible through the donation of microplastic particles, collected from Ontario lakes and rivers by Pollution Probe.  

Learn more in the published article, available here.