Optimized random forest method for predicting optical properties of a photonic crystal fiber

Document Type

Article

Publication Date

9-1-2026

Abstract

Conventional optimization of photonic crystal fiber (PCF) properties often requires extensive computational resources, making it inefficient for complex PCF structures necessitating multiple simulations. This study proposes a modified Random Forest (RF) model to predict key optical properties of solid-core PCFs, including effective refractive index, effective mode area, dispersion, and confinement loss. By optimizing hyperparameters, the modified RF model demonstrates superior predictive accuracy, reducing the mean squared error (MSE) by 60% and 50% compared to reported and modified artificial neural network (ANN) models, respectively. Notably, the modified RF model exhibits a unique ability to extrapolate beyond the training dataset, accurately predicting optical properties for PCF structures with geometric parameters outside the training range. This extrapolation capability, which is not reported in prior studies, is crucial for optimizing PCF properties and exploring new fiber designs without the need for computationally intensive numerical simulations. The modified RF model thus offers a powerful and efficient alternative to conventional methods for advancing PCF design and optimization.

Keywords

Extrapolation; Fibers; Forecasting; Mean square error; Neural networks; Optimization

Publication Title

Optical Fiber Technology

ISSN

1068-5200

DOI

10.1016/j.yofte.2026.104652

Volume

100

Share

COinS