Using machine learning techniques to improve quality of transmission estimation in multivendor optical open line systems

The adoption of Machine Learning techniques in optical communication is motivated by extraordinary growth of network traffic. Various machine learning models are used to predict the quality of transmission (QoT-E) of an unestablished light path and their performance is evaluated. Estimation of (QoT) is vital for diminishing provisioned margins and for optimizing design of optical network. It could be during design phase of network or in an already deployed network. In this thesis ML regression is investigated that predicts, whether the OSNR of unestablished light paths fulfills system requirements or not. It is trained and tested on synthetic data and its performance is assessed by using different algorithms and also by using different combinations of features. Some feature importance techniques are used for selection of features.