Comparing The Performance Of Machine Learning Methods In A Surrogate Approach To Optimize Low-Thrust Interplanetary Trajectories