Machine Learning Application to Transmission Quality Assessment in Optical Networks

This paper presents the application of 2 selected machine learning methods for transmission quality assessment in optical networks. The proposed solution uses real data to train classification models that predict whether a given optical channel meets transmission quality requirements. The training data is extracted directly from the running network through the control plane. Several attribute selection strategies have been analysed. The available attributes are divided into several subsets. The performance of the selected machine learning algorithms for each subset is compared using ROC and PR curves. The results are used to evaluate the predictive performance of the selected machine learning algorithms when applied to the optical channel classification task and to determine how this performance is related to the set of attributes used.