The ever-increasing amount of dynamic traffic with variations in both volume and service time will develop quickly in the foreseeable future, to a level that network link reconfigurations are required to support the bursting or abrupt change of network traffic. Optical networks need to be evolved to provide and manage network bandwidth automatically in an on-demand approach. In this project, machine-Learning (ML) technologies will be explored in dynamic optical networks to optimise network function and link bandwidths of optical networks and allocate network resources efficiently. ML-based network analytics and predictions will replace the timeconsuming method of manual link set up. A new mechanism of automatic resource provision and autonomous network management need to be developed to replace the set-and-forget operation model used in current optical networks.