This article proposes a technique to reduce data dimensionality in bioinformatics tasks. In the cases where there are thousands of attributes and only few hundred instances even the scalable classification methods benefit from diminishing the number of features. But at the moment there is no way of finding the best feature selection and evaluation method for a particular data set and the performance of these methods varies a lot depending on the specific nature of a data set. Therefore it is necessary to find a robust technique that would perform the process of feature selection without the impact of a specific method. The system proposed in this article minimizes this effect while still producing highly informative feature subsets that are easier to comprehend, interpret and use in classification.