PCA combined with SVM assisted fluorescence spectroscopy for classification of microplastics

Abstract

Microplastic (MP) pollution presents a significant challenge to environmental protection and requires rapid detection and classification methods. We utilize machine learning methods coupled with fluorescence spectroscopy detection to improve the accuracy of MP detection and classification. To comprehensively explore MP classification, Principal Component Analysis (PCA) and PCA-SVM methods are used to analyze 2400 spectral data samples of six types of MPs. Each MP category is divided into a training set comprising 200 spectra and a test set containing 200 spectra to ensure robust evaluation. The initial SVM model achieves 100% classification accuracy for the test set, the associated computational burden is significant, with a training time of 42.14 seconds and a prediction time of 8.23 seconds. To enhance efficiency, we integrate the PCA algorithm, which reduces feature dimensionality without compromising accuracy. The integration of PCA significantly reduces training time to 9.46 seconds and prediction time to 0.05 seconds while maintaining a 100% classification accuracy rate. These results highlight the efficacy of our methodology in efficiently classifying MPs. Combining machine learning and fluorescence spectroscopy, our research provides a promising solution to the pressing challenge of monitoring MP contamination.