Comparative Analysis of Machine Learning Algorithms for Wide-range Refractive Index Classification in Multimode Interference No-Core Fiber Sensors.

Abstract

Multimode interference (MMI) fiber sensors based on no-core fiber (NCF) are attracting growing interest due to their structural simplicity, cost-effectiveness, and ability to operate without cladding removal or chemical modifications. These sensors are widely utilized in applications requiring precise refractive index (RI) detection, such as environmental monitoring and biomedical diagnostics. However, accurate classification across a wide RI range is challenging, particularly due to the transition between guided modes (when the surrounding RI is below the fiber RI) and leaky modes (when the surrounding RI is above the fiber RI). This study proposes a machine learning (ML)-based approach to address these challenges and enhance RI classification accuracy. Transmission spectra from an NCF-based MMI sensor were acquired under both low-RI (LRI) and high-RI (HRI) conditions and analyzed using MATLAB. Three ML classifiers involved, Decision Tree (DT), Support Vector Machine (SVM), and Neural Network (NN) were evaluated using standard metrics: accuracy, precision, recall, and F1-score. The NN model achieved the highest classification accuracy at 80.0%, outperforming SVM (77.5%) and DT (42.5%). These results demonstrate the effectiveness of ML, particularly NN, in improving the performance and reliability of fiber optic sensors for broad-range RI detection.