A Hybrid Computational Framework for PCF‐based SPR Sensor With Machine Learning and xAI

Photonics crystal fiber (PCF)‐based sensors employing surface plasmon resonance (SPR) have become a highly effective tools for detecting molecular interactions without using any labels. This study introduces highly sensitive PCF‐based SPR sensor, utilizing rectangular air holes with traditional circular air holes, which achieves an impressive wavelength sensitivity of 15000 nm/RIU and an amplitude sensitivity of 1025.02 RIU −1 , spanning a broad refractive index (RI) range from 1.31 to 1.39. The proposed sensor was assessed by numerical analysis, utilizing the finite element method along with a perfectly matched layer. Furthermore, this research employs machine learning (ML), specifically random forest regressor (RFR), gradient boosting regressor (GBR), and categorical boosting regressor (CatBR), to predict confinement loss and sensitivity, marking a notable progression in reducing computational time (0.002 sec). An explainable AI (xAI) method is further applied to understand the dependency of the sensor parameters. These algorithms provide adaptive learning capabilities and meaningful data‐driven interpretation, resulting in an impressive R 2 value of 0.98, confirming their effectiveness in forecasting sensor performance. This integration of SPR, PCF, and ML not only enhances sensor optimization but also presents a rapid, efficient framework for performance estimation, emphasizing the significant potential of ML in the advancement of optical sensor development.