Visualization of training samples using top 3 selected features by different feature selection methods. The feature selection methods are: (a) BW, (b) ANMM without feature pre-selection and sample clustering, (c) ANMM. Results of MRMR were not listed due to space limitation. Figure 4 shows that MRMR did not perform better than BW on this data. In these figures, different marker types represent samples in different classes, and the mis-specifications are depicted with red edge. In (c) samples in different clusters are filled with different colors.