Replication Data for: Integrating C-H Information to Improve Machine Learning Classification Models for Microplastic Identification from Raman Spectra

The development of uniform, consistent spectroscopic databases of Raman spectra is important for the community to maximize the value of emerging machine learning techniques. This dataset contains processed and augmented Raman spectra acquired on a variety of common plastics, with variations in manufacturer and properties such as plastic color. The Raman spectra span the frequency window from 300 to 3900 cm-1, were collected using variations in instrumentation settings, were interpolated to 1 cm-1 wavenumber spacing to ensure compatibility, and were augmented 5X by random scaling and artificial noise introduction. Three different versions of the data are provided, each enabling exploration of a different strategy for training machine learning classification models. This data was used to train microplastic classification models using K-nearest neighbor algorithm of the sklearn package in python, as published in the associated manuscript. Python pickle files are included in the dataset, which contain the optimized models and supporting information for the models. The data are being posted in support of this research.

The data was created by the authors.