Orbit-Raising maneuvers for satellites utilizing solar electric propulsion allows for substantial mass, and thus cost, savings. Optimization of such trajectories is difficult and requires solving a non-linear, non-convex problem. A new emergent methodology breaks the overarching optimization problem into a sequence of smaller, easy to solve optimization sub-problems. This methodology can provide solutions in a fast and automated manner; however, these solutions fall short of the global optimal. Recognizing this shortcoming, this thesis investigates improving those solutions through the use of machine learning. Not only for typical Near-Earth missions, but also for missions that explore the Cis-Lunar realm. In this thesis, artificial neural networks are trained to make predictions for two metrics of interest. The first is the number of revolutions for the transfer and the second is the total amount of time the transfer takes. These predictions assist both direct optimization of the sub-optimal solutions that are generated and enables adaptive frameworks for the methodology itself. The first part of this research details recent improvements to the sequential methodology and how it is used to generate training data for the artificial neural networks. Two different transfer examples are presented, one for Near-Earth and a second for Cis-Lunar. The second part of this research discusses the results for the training of various artificial neural networks. Each with a different combination of inputs, hidden layers, and total neurons. Then the best performing architecture combination for each transfer example is selected and its performance over its respective training database is analyzed.