To this aim we took as input each learning data set and obtained its projection in the d_-dimensional space by means of six competing techniques: Isomap, L-IsomapMSC, L-IsomapSC and its unsupervised variant based on the spread coefficient (L-IsomapUSC), Landmark Isomap with random landmark selection (L-IsomapR) and based on the minimum spanning tree-cut algorithm (L-IsomapMST) described in [5]