Unraveling overoptimism and publication bias in ML-driven science
Fuente:
arXiv
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866909251336667136 |
|---|---|
| author | Saidi, Pouria Dasarathy, Gautam Berisha, Visar |
| author_facet | Saidi, Pouria Dasarathy, Gautam Berisha, Visar |
| contents | Machine Learning (ML) is increasingly used across many disciplines with impressive reported results. However, recent studies suggest published performance of ML models are often overoptimistic. Validity concerns are underscored by findings of an inverse relationship between sample size and reported accuracy in published ML models, contrasting with the theory of learning curves where accuracy should improve or remain stable with increasing sample size. This paper investigates factors contributing to overoptimism in ML-driven science, focusing on overfitting and publication bias. We introduce a novel stochastic model for observed accuracy, integrating parametric learning curves and the aforementioned biases. We construct an estimator that corrects for these biases in observed data. Theoretical and empirical results show that our framework can estimate the underlying learning curve, providing realistic performance assessments from published results. Applying the model to meta-analyses of classifications of neurological conditions, we estimate the inherent limits of ML-based prediction in each domain. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2405_14422 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Unraveling overoptimism and publication bias in ML-driven science Saidi, Pouria Dasarathy, Gautam Berisha, Visar Machine Learning Artificial Intelligence Computers and Society Machine Learning (ML) is increasingly used across many disciplines with impressive reported results. However, recent studies suggest published performance of ML models are often overoptimistic. Validity concerns are underscored by findings of an inverse relationship between sample size and reported accuracy in published ML models, contrasting with the theory of learning curves where accuracy should improve or remain stable with increasing sample size. This paper investigates factors contributing to overoptimism in ML-driven science, focusing on overfitting and publication bias. We introduce a novel stochastic model for observed accuracy, integrating parametric learning curves and the aforementioned biases. We construct an estimator that corrects for these biases in observed data. Theoretical and empirical results show that our framework can estimate the underlying learning curve, providing realistic performance assessments from published results. Applying the model to meta-analyses of classifications of neurological conditions, we estimate the inherent limits of ML-based prediction in each domain. |
| title | Unraveling overoptimism and publication bias in ML-driven science |
| topic | Machine Learning Artificial Intelligence Computers and Society |
| url | https://arxiv.org/abs/2405.14422 |