An apparatus and method for optimizing dynamically the performance of an optical network, said apparatus comprising at least one learning engine adapted to update a learning model in response to network metrics of said optical network collected during operation of said optical network, wherein the updated learning model is used to generate channel rank information for network channels; and a recommendation engine adapted to change a network channel throughput, a signal path and/or a spectral location of at least one network channel based on the channel rank information generated by the learning model of said learning engine.