Question about reproducing results in Table 3
Question about reproducing results in Table 3
Thank you to the authors for this work. I have a question regarding the reproducibility of the results reported in Table 3. In the original paper, it appears that when oversampling is applied—as described by Gupta et al.—the oversampling step occurs before splitting the data into test, train, and validation sets. This approach could result in identical samples appearing in all three sets, potentially leading to data leakage (and it will increase the AUROC because some oversampling samples can exist in both test/train). When I try to turn off Oversampling, the AUROC drop significantly.
Could you please clarify how your group handles this situation? Did you use the same Oversampling approach or different one? Thank you very much.
Hello
@HugoLina2025
,
We used the same oversampling method as the original authors to reproduce their experiments. The oversampling does indeed occur before the split. However, we did try multiples experimental setups with and without oversampling, but it did not affect significantly the AUROC score as you mentioned.
I hope this helps,
Kind regards
Thank you for your response
Best regards,