Figure 2

The masking procedure. For clarity only four predictors are drawn. The orange elements represent predicted TM segments. It is assumed in this example that three overlapping TM segments need to intersect with the voting window to have a positive consensus voting (i. e. the SVM model predicts a TM segment). (A) Only two segments are in the voting window, thus the voting result is negative. (B) Due to the negative voting result, the first segment in the window is masked and not used for further prediction anymore. The newly applied window contains three overlapping segments. (C) The consensus is reached in favor of the TM segment and therefore all overlapping segments are masked. Note that if both segments overlapping with the first window (the one displayed in (A)) had been excluded, there would not be any resulting consensus TM segments.