Modern decision support systems make use of machine learning and artificial intelligence to solve complicated problems. One of them is classification, understood in this context as assigning objects to categories. Amongst many methods to achieve this goal, rulebased systems pay special attention, because they provide an end-user not only with direct answers to a given problem, but also produce useful insights into correlations present in a dataset. In this article new method has been proposed − application and modification of Leo Breiman’s original Random Forest solution combined with backwards elimination (known from classic regression) − and tested on real credit decisions dataset. Differences in classification metrics between base and augmented classifier were checked using cross-validation testing, and statistical significance. The article concludes with further research suggestions.