A graph classification approach to determine when to decompose optimization problems

Decomposition-based optimization algorithms are widely used for the solution of optimization problems. However, the choice of whether a decomposition-based solution approach should be selected over a monolithic one is not apparent in general. In this work, we propose a graph classification approach for determining a-priori when to use a decomposition-based solution approach for the solution of convex mixed integer nonlinear optimization problems. We apply the proposed approach to benchmark optimization problems and analyze the predictive performance of the classifier.