Advanced PCF-SPR biosensor design and performance optimization using machine learning techniques

Photonic Crystal Fiber-Surface Plasmon Resonance (PCF-SPR) is an advanced optical biosensing technology that utilizes specialized fibers to detect refractive index (RI) changes caused by molecular interactions. This study proposed a dual-core, gold-coated PCF-SPR biosensor designed to operate across analyte RIs ranging from 1.31 to 1.40 and a broad wavelength range of 0.40 µm to 0.90 µm, with the integration of machine learning (ML) techniques to enhance performance optimization and improve predictive accuracy. Using the COMSOL Multiphysics Simulator, we generated a dataset of 1868 samples derived from our novel sensor model. The sensor achieved exceptional performance, reaching wavelength sensitivity (S_λ) of up to 9000 nm/RIU, amplitude sensitivity (S_A) of −1141.93 RIU⁻¹, and resolution (R) as high as 1.11 × 10⁻⁵ RIU. Numerous machine learning (ML) models were employed to predict key parameters, including the effective refractive index (N_eff) and confinement loss. Among these, the Random Forest Regressor (RFR) achieved outstanding results, with an R-squared value of 0.9997, a minimal mean absolute error (MAE) of 4.51 × 10⁻⁴, and a mean squared error (MSE) of 8 × 10⁻⁶ for N_eff prediction. RFR also excelled in superior accuracy in predicting confinement loss. With its exceptional sensitivity and detection capabilities, the proposed sensor offers significant potential for biological applications. Future work will integrate Explainable AI (XAI) to identify critical features, driving further advancements in sensor performance and precision.