CACTUS Improves Feature Stability Under Missingness
Paulina Tworek et al. (submitted 19 Feb 2026) present CACTUS, an explainable machine learning framework addressing robustness, interpretability, and feature instability in small, incomplete clinical datasets. They evaluate CACTUS on a 568-patient haematuria cohort for bladder cancer, benchmarking against random forests and gradient boosting with controlled missingness. CACTUS matches or exceeds predictive performance while preserving markedly higher top-feature stability as data completeness degrades, aiding trustworthy decision support.
Key Points
- 1Introduces CACTUS, an explainable ML framework quantifying feature stability under varying missingness levels.
- 2Demonstrates higher top-feature stability versus random forests and gradient boosting on 568-patient haematuria cohort.
- 3Enables more trustworthy clinical models by prioritizing interpretable, stable features under realistic data degradation.
Scoring Rationale
Novel, actionable framework with strong relevance; limited by single-cohort arXiv preprint evaluation pending peer review.
Sources
Public references used for this report.
Practice with real Health & Insurance data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card
See all Health & Insurance problems
