Abstract: Systems and methods for optical amplifier failure prediction using Machine Learning (ML), such as for an Erbium-Doped Fiber Amplifier (EDFA), are described. A method include obtaining a plurality of inputs from an optical amplifier associated with an optical network; analyzing the plurality of inputs with a trained machine learning model; obtaining an estimate of a total pump current of the optical amplifier as an output of the trained machine learning model; and comparing the estimate of a total pump current to a measured total pump current of the optical amplifier.