A Machine Learning Approach for Efficiently Predicting Polymer Aging from UV-Vis Spectra.

Area of Science:

• Polymer Science
• Spectroscopy
• Machine Learning

Background:

• Traditional ultraviolet-visible (UV-Vis) spectroscopy for polymer analysis is precise but costly and inefficient.
• Assessing polymer solution aging is crucial for material quality control and performance prediction.

Purpose of the Study:

• To develop an innovative and cost-effective approach for forecasting UV-Vis spectral alterations in polymer solutions due to aging.
• To utilize readily accessible feature descriptors and machine learning (ML) algorithms for this prediction.

Main Methods:

• Input features included wavelength and blue (B), green (G), and red (R) color values of polymer solutions.
• Seven classical ML models were trained and evaluated using 10-fold cross-validation.
• Model performance was assessed using metrics like Pearson correlation coefficient (r) and Mean Absolute Error (MAE).

Main Results:

• All seven ML models demonstrated excellent performance in predicting spectral changes.
• The ExtraTree model exhibited superior predictive ability, achieving an r of 0.9859 and an MAE of 0.0457.
• The proposed method provides a rapid and economical alternative for evaluating polymer solution aging.

Conclusions:

• The developed ML approach effectively predicts polymer solution aging using simple color and wavelength data.
• This method offers a practical, rapid, and cost-effective solution compared to traditional spectroscopic techniques.
• The findings support the use of ML in material science for efficient property evaluation.