A Principal Component Analysis-Integrated Machine Learning Approach for Predicting Gas-Phase VUV/UV Absorption...

Area of Science:

• Computational chemistry
• Spectroscopy
• Machine learning

Background:

• Predicting gas-phase vacuum ultraviolet/ultraviolet (VUV/UV) absorption spectra is crucial for chemical research.
• Standalone random forest (RF) models struggle with high-dimensional spectral data, failing to capture interwavelength correlations.
• Previous methods using molecular descriptors and RF models had limitations in spectral data analysis.

Purpose of the Study:

• To develop an enhanced machine learning framework for predicting VUV/UV absorption spectra.
• To overcome the limitations of standalone RF models in handling high-dimensional spectral data.
• To improve the accuracy and efficiency of spectral prediction models.

Main Methods:

• Integration of principal component analysis (PCA) for dimensionality reduction with random forest (RF) regression.
• Application of PCA to spectral data, retaining 99% of variability.
• Training an RF regressor on PCA-transformed spectral data and molecular structure-based descriptors.

Main Results:

• The PCA-integrated RF model achieved equivalent or better R² values (up to 0.727) compared to previous methods.
• Training time was reduced by over 15-fold due to PCA dimensionality reduction.
• Weighted PCA revealed that principal components corresponded to chemically meaningful spectral regions, enhancing interpretability.

Conclusions:

• PCA is a valuable tool for dimensionality reduction in spectral data analysis.
• Integrating PCA with RF models enhances the prediction of VUV/UV absorption spectra.
• The developed framework offers a more interpretable and scalable approach to spectral prediction and chemical structure analysis.