Abstract
Photonic crystal fiber-based surface plasmon resonance (PCF-SPR) biosensors are widely recognized for their high sensitivity in medical diagnostics. Most existing studies use iterative simulations, which are time-consuming and not suitable for efficient, scalable design. This study presents a highly sensitive PCF-SPR biosensor with a comparatively simple design integrating machine learning (ML). The structure was designed and evaluated using COMSOL Multiphysics simulation software across a refractive index (RI) range of 1.33 to 1.40. Subsequently, a simulated dataset was generated, and several ML models were applied to it to explore optimal design parameters. Model performance was evaluated using mean absolute error (MAE), mean squared error (MSE), root mean squared error (RMSE), and R-squared (R2). The proposed sensor achieved a high wavelength sensitivity of 7000 nm/RIU for RI values 1.36 to 1.38 (basal cancer cells) and 10,000 nm/RIU for RI values between 1.33 and 1.40. Among all models, Gradient Boosting Regression (GBR) performed best in predicting the real part of the effective refractive index (MAE: 8 × 10−4, MSE: 1.4 × 10−7, RMSE: 3 × 10−4, R2: 0.9912). In contrast, Extreme Gradient Boosting Regression (XGBR) was most effective for predicting confinement loss (MAE: 0.467, MSE: 1.4661, RMSE: 1.1896, R2: 0.9278). These models provided data-driven insights that enhanced sensor performance, while the integration of ML with simulation reduced computational time and improved the overall design process. The proposed biosensor outperforms previously reported PCF-SPR sensors, particularly in detecting basal cancer cells.