Fig. 21

Comparison between (i) the EWs in our paper using the ERT algorithm, (ii) the EWs using the SPA algorithm proposed by Gu et al. on our dataset, and (iii) the EWs using SPA, GA, and BRT as in [19]. The vertical dashed lines represent the bands chosen by SPA, GA, and BRT in Gu’s work, with each point annotated by the percentage overlap with ERT-selected bands across the four experiments. The purple shading indicates the number of ERT dataset splits in which a band was selected, with darker shades representing higher frequencies of selection