.png){kind=icon}
.png){kind=icon}
Return
Decision tree learning algorithms for WDM systems’ performance prediction
.png){kind=icon}
PRE
AI
DOI：10.1016/j.optlastec.2025.114217.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
We consider the problem of predicting the performance of an operational wavelength division multiplexing (WDM) system using machine learning techniques. We assess the performance of the WDM system using the Quality Factor metric (QF). To that end, we construct an actual WDM dataset, whose records are generated using four live WDM systems. We export these records using different optical measurement tools, such as reflectometers, spectrum analyzers, and the operating management system for each WDM system. To circumvent the difficulty of theoretical analysis of WDM system performance, we propose using machine learning techniques to predict such a performance. We employ the decision tree learning techniques to predict the QF, namely, the decision tree regressor (DTR) and the decision tree classifier (DTC). By training DTR on the constructed dataset and using shuffle-split cross-validation, we show that DTR outperforms DTC in terms of the R2 and F1 score, specifically; our DTR achieves an R2 score of 96.2%, while the testing score records 89.65%. Moreover, the DTR decreases the average prediction error by approximately 9.9% compared to the DTC, while also achieving a 5.1% improvement in minimizing larger prediction errors relative to the DTC. The proposed DTR model surpasses the established benchmarks of Support Vector Machine (SVM) and Extreme Gradient Boosting (XGBoost). SVM achieves a training score of 92.23% and a testing score of 89.32%, while XGBoost classifier demonstrates training and testing scores of 90.98% and 85.58%, respectively, and an XGBoost regressor test score of 99.6%. In addition, our numerical results show ≈ 3000 reduction in total running time when using DTR compared to SVM.
Journal
O
IF:
0
Papers:
880
Citations:
0