Machine learning-assisted optimization of a BK7/Ag/As₂S₃/graphene surface plasmon resonance biosensor for high-sensitivity urine glucose detection.

Blood glucose monitoring remains challenging due to the invasiveness of finger-prick methods, enzyme instability in electrochemical sensors, and the limitations of laboratory-based techniques for continuous real-time detection. This study presents a theoretically optimized multilayer surface plasmon resonance (SPR) biosensor based on a BK7 prism/silver/arsenic trisulfide/graphene structure operating at 633 nm for non-invasive glucose detection in urine. The novelty of the proposed approach lies in the integration of an ultrathin As₂S₃ dielectric spacer and graphene sensing layer with a systematic electromagnetic optimization and Gradient Boosting Regressor (GBR)-based predictive framework within a single SPR sensing platform. Using the transfer matrix method and electromagnetic field analysis, the optimized thicknesses were identified as 52 nm for silver, 1.0 nm for arsenic trisulfide, and 1.32 nm for graphene, enabling strong plasmonic coupling and enhanced sensing performance. The proposed sensor achieved sensitivities ranging from 33.333 to 500°/RIU, a maximum figure of merit of 555.556 RIU-1, and a detection limit of 0.001 RIU over a refractive index range of 1.335-1.347. Furthermore, the combination of multilayer optical optimization with GBR-based surrogate modelling provides an efficient approach for predicting the resonance response across the investigated design space and reducing the need for repeated electromagnetic simulations. Machine learning validation using a Gradient Boosting Regressor achieved R2 values above 0.99, confirming the robustness and reliability of the proposed biosensor design.