Beyond Traditional airPLS: Improved Baseline Removal in SERS with Parameter-Focused Optimization and Prediction

Area of Science:

• Spectroscopy
• Analytical Chemistry
• Computational Chemistry

Background:

• Baseline correction is crucial for accurate Raman and surface-enhanced Raman spectroscopy (SERS).
• The adaptive iterative reweighted penalized least-squares (airPLS) method faces challenges with parameter sensitivity and complex spectra.

Purpose of the Study:

• To develop an optimized airPLS (OP-airPLS) algorithm for improved baseline correction.
• To implement a machine learning model for predicting optimal airPLS parameters.
• To enhance the accuracy and efficiency of spectral preprocessing in Raman spectroscopy.

Main Methods:

• Systematic fine-tuning of airPLS parameters using adaptive grid search.
• Development of a machine learning model for parameter prediction via spectral shape recognition.
• Evaluation using 6000 simulated spectra across 12 spectral shapes.
• Implementation of a principal component analysis and random forest (PCA-RF) model for direct parameter prediction.

Main Results:

• OP-airPLS achieved an average percentage improvement (PI) of 96 ± 2% in baseline correction accuracy.
• Maximum MAE reduction by 99.46 ± 0.06%, minimum MAE reduction by 91 ± 7%.
• PCA-RF model demonstrated robust performance with 90 ± 10% PI, processing spectra in 0.038 s.
• Optimal parameters for spectral shapes were found in a well-defined linear region.

Conclusions:

• OP-airPLS significantly enhances baseline correction accuracy but requires computational resources and known baselines.
• The PCA-RF model offers a computationally efficient alternative for predicting optimal parameters directly from spectra.
• Performance of the PCA-RF model on real spectra depends on signal-to-noise ratio and spectral similarity to training data.