Integrating C-H Information to Improve Machine Learning Classification Models for Microplastic Identification from...

Area of Science:

• Environmental Science
• Analytical Chemistry
• Data Science

Background:

• Microplastics are widespread environmental pollutants requiring complex analysis.
• Current machine learning (ML) models for microplastic identification from Raman spectra are limited to the fingerprint region.
• Diverse plastic chemistries pose challenges for accurate identification.

Purpose of the Study:

• To improve ML classification models for microplastic identification.
• To explore the utility of higher frequency Raman spectral regions (2500-3600 cm⁻¹).
• To investigate data acquisition and analysis strategies for enhanced microplastic detection.

Main Methods:

• Employed the k-nearest neighbor (k-NN) algorithm for classification.
• Included higher frequency Raman spectral data (C-H bond region) alongside the fingerprint region.
• Tested localized spectral normalization strategies for independent data acquisition.

Main Results:

• Incorporating the C-H bond region (2500-3600 cm⁻¹) significantly improved ML model performance.
• Enhanced identification capabilities for plastics difficult to distinguish using only the fingerprint region (e.g., ABS, PVC, PU, POM).
• Localized normalization of independently acquired spectra proved effective and practical.

Conclusions:

• Higher frequency Raman spectral regions offer valuable information for microplastic identification.
• Combining spectral regions and employing localized normalization enhances ML model accuracy.
• This approach offers a more robust and efficient method for microplastic analysis in environmental samples.