Customizable Machine-Learning Models for Rapid Microplastic Identification Using Raman Microscopy.

Area of Science:

• Environmental Science
• Analytical Chemistry
• Spectroscopy

Background:

• Raman spectroscopy is vital for microplastic identification.
• Equipment variations lead to inconsistent data, hindering collaborative research.
• A need exists for standardized, adaptable analytical tools.

Purpose of the Study:

• To develop a strategy for overcoming data inconsistencies in Raman spectroscopy for microplastics.
• To create robust and customizable machine-learning classification models.
• To demonstrate the adaptability of these models across different spectrometer setups.

Main Methods:

• Compiled a database of high-resolution, full-window Raman spectra.
• Developed machine-learning models using random-forest, K-nearest neighbors, and multi-layer perceptron algorithms.
• Tested model accuracy with downgraded spectral data and non-ideal experimental conditions.

Main Results:

• Achieved >95% classification accuracy for microplastics.
• Accuracy was maintained with spectral data downgraded to 1, 2, 4, or 8 cm^-1 spacings.
• Models performed well under non-ideal conditions, including high sampling rates and out-of-focus particles.

Conclusions:

• A novel strategy using a spectral database enables creation of robust, adaptable microplastic classification models.
• The approach overcomes spectrometer variations and enhances the development of communal analytical tools.
• This method supports accurate microplastic identification in diverse and challenging environments.