Smart optical biosensor for edible oil detection with machine learning integration

A photonic crystal fiber (PCF) biosensor is an advanced optical sensing device that can detect material with high sensitivity. This study presents a heptagonal core-shaped PCF biosensor with a circular core for highly sensitive detection of edible oils in the terahertz (THz) frequency range. Numerical analysis is performed using the finite element method based on Maxwell's equations in COMSOL Multiphysics software. The sensor performs over a frequency range of 1.0 THz to 3.0 THz. The sensing properties are evaluated for four edible oils, such as coconut, olive, mustard, and sunflower. Key optical properties, such as relative sensitivity (RS), effective material loss (EML), effective mode area, effective mode index, birefringence, and total power fraction, are calculated, analyzed, and predicted. The optimized structural parameters give superior performance with a maximum RS of 98.14% and a minimum EML of 0.0046758 cm^-1 at the operating frequency of 2.2 THz. To enhance computational efficiency, a machine learning (ML)-based predictive framework is developed using a dataset of 3024 samples collected from parametric simulations of the following software. Several regression models as Artificial Neural Network (ANN), Support Vector Regression, Random Forest, Extreme Gradient Boosting and Bagging SVR are analyzed. But the ANN model demonstrates the best performance. The ANN shows an average coefficient of determination (R²) of 0.9781 and a mean absolute error (MAE) of 0.0636. The trained ANN predicts the following optical properties accurately. It reduces time for numerical simulations. The integration of FEM-based modeling with ML-driven prediction establishes an efficient framework for rapid sensor optimization.