Active Wavelength Load as a Feature for QoT Estimation Based on Support Vector Machine

Reconfiguration procedures in optical transmission systems assisted with artificial intelligence (Al) present an innovative approach towards the mitigation of network resources mismanagement. Regression and classification tools have been studied in recent years with the aim to predict performance metrics such as Bit Error Rate (BER) and Optical Signal-to-Noise Ratio (OSNR). We have generated synthetic OSNR labeled data, which has been used for training a Support Vector Machine (SVM) classifier, in order to predict the OSNR performance upon provisioning a wavelength channel (lightpath). Information on the active lightpaths in the network is used to train the learning model, together with network topology configuration features. Our results demonstrate a 96.2% multi-class classification accuracy to predict QoT of unestablished lightpaths in topology independent (generic) scenarios.