PreOBP_ML: Machine Learning Algorithms for Prediction of Optical Biosensor Parameters.

Area of Science:

• Photonics and Biosensing
• Computational Modeling

Background:

• Optical biosensor development relies heavily on time-consuming simulations.
• Key performance metrics like effective index and power fraction are critical for sensor evaluation.

Purpose of the Study:

• To explore machine learning (ML) approaches for accelerating optical biosensor design.
• To predict crucial optical sensor parameters using ML models.
• To compare the efficacy of various regression techniques for this application.

Main Methods:

• Applied machine learning regression models: Least Squares (LS), LASSO, Elastic-Net (ENet), and Bayesian Ridge Regression (BRR).
• Utilized COMSOL Multiphysics simulation data with core radius, cladding radius, pitch, analyte, and wavelength as input vectors.
• Performed comparative analysis of model performance using R2-score, Mean Average Error (MAE), and Mean Squared Error (MSE).

Main Results:

• All ML models achieved an R2-score exceeding 0.99, demonstrating high prediction accuracy.
• The developed models predicted optical biosensor parameters with a design error rate below 3%.
• Sensitivity, power fraction, and confinement loss were analyzed using predicted and simulated data.

Conclusions:

• Machine learning offers a viable and efficient alternative to traditional simulation methods for optical biosensor development.
• The study validates the use of ML for accurate prediction of optical sensor performance metrics.
• This research facilitates faster optimization and improved design of optical biosensors through ML-based approaches.