When perfect is a warning: explanations expose shortcuts that accuracy hides in cancer gene-expression classifiers
Hamail Ayaz, Pooja Verma, Muhammad Mahmood Ali, OLADOSU OLADIMEJI, Ayodeji Olusegun Ibitoye
The paper shows cancer classifiers can achieve perfect accuracy by learning data source shortcuts rather than biological features, and introduces a method to detect and mitigate this issue.
bioRxiv · Nov 3, 2026 · PreprintBioinformatics