Sandip Das
Abstract An artificial intelligence (AI)-assisted multi-objective optimization framework is presented for the systematic enhancement of a dual-channel photonic crystal fibre surface plasmon resonance biosensor designed for early-stage jaundice detection. The proposed structure employs symmetric elliptical analyte channels integrated within a microstructured cladding and coated with a gold–graphene bilayer to promote strong plasmonic confinement and enhanced biomolecular interaction. Finite element method (FEM) simulations were performed over a physiologically relevant refractive index range (1.333–1.365) corresponding to urine and serum samples with varying bilirubin concentrations. To overcome the limitations of exhaustive parametric scanning, a machine-learning framework combining Gaussian process surrogate modelling, SHapley Additive exPlanations-based interpretability analysis, and Non-dominated Sorting Genetic Algorithm II multi-objective optimization was implemented to explore the high-dimensional design space. The optimized configuration exhibits a FEM-validated wavelength sensitivity of 1892 nm RIU − 1 , compared to 1200 nm RIU − 1 and 600 nm RIU − 1 for urine and serum in the baseline design. The enhanced structure further achieves a figure of merit approaching 41 RIU − 1 with near-perfect linearity ( R 2 = 0.999 994 ). The reported performance improvement originates from optimized geometric parameters identified through the AI-assisted framework, rather than from the AI model itself. By enabling efficient exploration of the high-dimensional design space, the proposed AI-physics integration yields over 57 % sensitivity enhancement for urine samples and more than a threefold improvement for serum conditions, while preserving spectral sharpness. These findings demonstrate a scalable and interpretable optimization strategy for high-performance plasmonic biosensors, establishing a systematic pathway for AI-guided photonic device engineering.