Implementation of Hybrid Machine Learning on Photonic Crystal Fiber Based Biosensor

In this article, a machine learning (ML) regression approach is proposed for detecting the core loss (CL) and effective refractive index (ERI) of the core mode of photonic crystal fiber (PCF) based biosensor. First, a PCF structure with hexagonal air hole pattern is considered as a waveguide for optical transmission in the near infrared region. The dataset for core mode analysis is generated through finite element method with surface plasmon resonance (SPR) principle. The highest wavelength sensitivity of 11000 nm/RIU has been observed. The ML regression algorithms like K-Nearest Neighbor (KNN), Extreme Gradient Boosting (XGBoost), Support Vector Regression (SVR) and a hybrid 1D-convolutional neural network (1D-CNN) & XGBoost is implemented for predicting the CL and ERI of the core modes. The proposed algorithms showed very high accuracy with Mean Squared Error (MSE) of 0.10432 for the hybrid regression model. A hybrid regression model showed almost the same wavelength sensitivity when comparing with simulated values. The proposed model helps to reduce the sources and time to find out the core mode analysis of PCF-SPR sensors. This ML regression model can be used for different types of PCF structures and also be used to optimize the design parameters suitable for biomedical applications.