Training and evaluating machine learning algorithms for ocean microplastics classification through vibrational...

Area of Science:

• Environmental Science
• Analytical Chemistry
• Data Science

Background:

• Microplastics are pervasive environmental contaminants with potential human health implications.
• Accurate characterization of microplastics is crucial for assessing risks and developing mitigation strategies.
• Fourier-transform infrared (FTIR) spectroscopy is a key technique for microplastic identification.

Purpose of the Study:

• To evaluate and compare machine learning classification algorithms for identifying microplastics using FTIR spectra.
• To assess the predictive performance of different models for 13 distinct polymer classes.
• To develop a robust and unbiased methodology for automated microplastic characterization.

Main Methods:

• Application of machine learning classification algorithms to FTIR spectra of marine microplastics.
• Rigorous pipeline methodology to ensure unbiased training and model selection.
• Utilized an oversampling technique to address dataset imbalance.
• Employed log-loss as the performance evaluation metric.

Main Results:

• Support Vector Machine (SVM) classifier demonstrated a favorable balance between simplicity and performance.
• The study successfully evaluated prediction performance across 13 polymer classes.
• The applied methodology ensured reliable and unbiased classification results.

Conclusions:

• Machine learning, particularly SVM, provides a fast and effective tool for automatic microplastic characterization from FTIR data.
• Accurate microplastic identification is essential for understanding their environmental and health impacts.
• This approach supports targeted actions to reduce the harmful effects of microplastic pollution.