Intelligent 18-Tupling Optical Links: PSO-Tuned ML for Predicting and Minimizing Signal Penalties

A novel Machine Learning (ML) framework is proposed, in this paper to predict and optimize key performance indicators in an 18-tupling optical communication system. It starts with delivering and training on a public 24,572-row OptiSystem dataset along with an open-source Python package, tupling_ml, containing rigorous preprocessing that includes correlation-based feature selection and dimensionality reduction through Principal Components Analysis (PCA). This procedure reduces the feature space from 68 to 4 dimensions while preserving over 80% of the variance. Several multi-output regression models are then evaluated, including multioutput Linear Regression (LR), Random Forest (RF), K-Nearest Neighbors (KNN), and a Regressor Chain of RF (RCRF). Validation is performed by means of 5-fold cross-validation (CV). The RCRF approach produces the best results, with a Mean Absolute Error (MAE) of 0.476, Mean Squared Error (MSE) of 0.963, and a coefficient of determination (<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$R^{2}$</tex-math></inline-formula>) of 0.893 on the test set. Finally, Particle Swarm Optimization (PSO) is employed to fine-tune two critical hyperparameters of the RF regressor: number of estimators and maximum depth, thereby further enhancing model performance by 3.3% better than the untuned RF chain. By enabling proactive resource allocation and lowering signal penalties, the proposed framework expedites the deployment of next-generation (<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$&gt;$</tex-math></inline-formula> 100Gb/s) optical fronthaul and fixed-wireless backhaul links.