Exploiting the ensemble paradigm for stable feature selection: A case study on high-dimensional genomic data