This paper explores a scheduling problem where the incoming material must be processed in batches and the number of batches required dominates the objective. A Mixed Integer Linear Program (MILP) model is presented which selects a portion of the available waste containers to process and groups them into processing batches. A machine learning approach was developed to improve optimization performance and extract useful knowledge by examining the optimization results for trends regarding which containers are selected. The learning process captured information about the containers that are usually selected. The optimization performance was not uniformly improved through biasing, indicating that simple biasing is not always effective. Additionally, an on-line method for determining which containers should be processed is described. The learning provides a mechanism for abstracting complex calculations into knowledge relevant to other organizational functions. This technique is applicable to selection-type problems where changes to the input data correspond to small changes in the optimization formulation.