If we use a means for non-deterministic parameter space segmentation, then every time we run TClass we are going to get a different set of features. Hence we can vote these results. As discussed in Section 5.4.1, this be combined with boosting and/or bagging.