When facing tight deadlines and rigorous quality assurance requirements, the manual refinement of test suites can offer testers a challenging assignment: analyzing and understanding large test suites, many times not even based on any specific rationale, and identifying their need to be simplified, transformed, or augmented. In order to help the tester to perform this task, Melba was introduced as a semi-automated method based on Machine Learning and Category-partition testing, aiming at helping the tester not only to understand test suites but also to improve them. However, previous studies have shown the need for stronger automation of the whole Melba process. In response to this need, we present here a study of the method along with improvements applied, in addition to the design and implementation of the Melba tool, developed in order to make test suites refinement with the Melba method an automated and straightforward activity.