Accuracy Enhancement in Refractive Index Sensing via Full-Spectrum Machine Learning Modeling.

Area of Science:

• Nanophotonics
• Machine Learning
• Spectroscopy

Background:

• Refractive index sensing is crucial for biosensing applications.
• Meta-grating structures offer tunable optical properties for sensing.
• Machine learning can enhance the analysis of complex spectral data.

Purpose of the Study:

• To develop and evaluate a full-spectrum machine learning framework for refractive index sensing.
• To compare the performance of titanium and silicon nanorod meta-gratings for sensing.
• To investigate the impact of spectral features on model accuracy.

Main Methods:

• Simulated absorption spectra from titanium and silicon nanorod meta-gratings.
• Extraction of 80 principal components from spectral data.
• Application of linear regression and five-fold cross-validation.
• Analysis of TE and TM polarized light.

Main Results:

• Titanium nanorods demonstrated significantly higher accuracy (up to 8128-fold improvement) due to broadband intensity changes.
• Silicon nanorods showed more limited gains because of spectral nonlinearity.
• Full-spectrum linear models outperformed single-feature models, especially for intensity-modulated sensors.
• Data-driven analysis identified optimal single-wavelength predictors.

Conclusions:

• Full-spectrum machine learning is effective for refractive index sensing.
• Titanium nanostructures are highly promising for advanced biosensing applications.
• Spectral shape and linearity significantly influence the performance of machine learning models in sensing.