Abstract
In contrast to typical optical fiber, photonic crystal fiber (PCF) exhibits a variety of unique optical properties as a result of its flexible cladding distribution. Nonetheless, assessing PCF optical characteristics becomes difficult when structural parameters fluctuate. This issue is a serious impediment to fully understanding and leveraging PCF's potential for diverse optical applications. Furthermore, the variety in structural factors makes it difficult to ensure PCF's consistent and reliable performance in practical optical systems. Artificial neural networks (ANN) are widely used to forecast the optical parameters of PCF. However, ANNs have issues when dealing with local minima. In contrast, solutions obtained from support vector machines regressions (SVM/SVR), Gaussian process regressions (GPR), and k-nearest neighbors regression (KNNR) are globally unique and avoid the dangers of slipping into local minimum values. Major properties such as effective refractive index (n
e
f
f
), confinement loss (α
c
) and dispersion (D) of photonic crystal fiber (PCF) were predicted using SVM/SVR, GPR, KNNR, random forest regression (RFR), gradient boosting regression (GBR), and ANN. To evaluate the performance of various regression algorithms, we created a database of 2912 samples including the X and Y directions. In terms of prediction accuracy and stability, SVM and GPR outperform other approaches.