The multilevel optimization approach, a strategy for solving large combinatorial optimization problems, involves a process of coarsening a large problem instance into a smaller, more manageable version, solving this simplified instance, and then projecting the solution back to the original problem instance. We propose a two level version of MLO enhanced by integrating machine learning techniques to improve the coarsening process. We demonstrate the efficacy of this approach in addressing a large location allocation problem: the demand maximizing battery swapping station location problem. Our results illustrate that leveraging machine learning to enhance coarsening can improve solution quality by up to 5% and can also reduce computational time, particularly for very large problem instances.