Skip to main content

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

Accurate data-driven prediction does not mean high reproducibility

Nature Machine Intelligence volume 2, pages 13–15 (2020)Cite this article

Subjects

A valid machine model is predictive, but a predictive model may not be valid. The gap between these two can be larger than many practitioners may expect.

Low reproducibility is a major concern in data-driven discovery with machine learning. Reproducibility requires that the relationships discovered from data are valid — that is, they are causal and true in a real system. A model that can provide accurate prediction of an outcome does not mean that the predictors used by the model are likely to be causes of the outcome. It is generally understood that association does not necessarily indicate causation. However, since causes can be used to make quality predictions, many practitioners take prediction accuracy as an indicator of how likely that a predictor is a cause of the outcome. In fact, prediction accuracy and causal validity are measures in two different worlds, and a wrong link between them is very harmful for data-driven discovery. This Comment discusses the reason of low reproducibility of data-driven discovery and provides practical guidelines for using general machine learning methods for data-driven discovery.

This is a preview of subscription content, access via your institution

Relevant articles

Open Access articles citing this article.

Access options

Buy article

Get time limited or full article access on ReadCube.

$32.00

Buy

All prices are NET prices.

Fig. 1: Accuracy versus validity.

References

  1. Nat. Genet. 51, 1 (2019).

  2. Runge, J. et al. Nat. Commun. 10, 2553 (2019).

    Article  Google Scholar 

  3. Hussein, A. A. et al. Br. J. Cancer 119, 724–736 (2018).

    Article  Google Scholar 

  4. Tam, V. et al. Nat. Rev. Genet. 20, 467–484 (2019).

    Article  Google Scholar 

  5. Lewis, R. A., Rao, J. M. & Reiley, D. H. in Proc. 20th International Conference on World Wide Web 157–166 (ACM, 2011).

  6. Pearl, J. Causality: Models, Reasoning, and Inference (Cambridge Univ. Press, 2009).

  7. Imbens, G. W. & Rubin, D. B. Causal Inference for Statistics, Social, and Biomedical Sciences: An Introduction (Cambridge Univ. Press, 2015).

  8. Reichstein, M. et al. Nature 566, 195–204 (2019).

    Article  Google Scholar 

  9. Hill, A. B. Proc. R. Soc. Med. 58, 295–300 (1965).

    Google Scholar 

  10. Pearl, J. Commun. ACM 62, 54–60 (2019).

    Article  Google Scholar 

Download references

Acknowledgements

This work has been supported by ARC Discovery Project grant DP170101306 and NHMRC grant 1123042.

Author information

Authors and Affiliations

  1. School of Information Technology and Mathematical Sciences, University of South Australia, Adelaide, South Australia, Australia

    Jiuyong Li, Lin Liu, Thuc Duy Le & Jixue Liu

Authors
  1. Jiuyong Li
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. Lin Liu
    View author publications

    You can also search for this author in PubMed Google Scholar

  3. Thuc Duy Le
    View author publications

    You can also search for this author in PubMed Google Scholar

  4. Jixue Liu
    View author publications

    You can also search for this author in PubMed Google Scholar

Corresponding author

Correspondence to Jiuyong Li.

Ethics declarations

Competing interests

The authors declare no competing interests.

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Li, J., Liu, L., Le, T.D. et al. Accurate data-driven prediction does not mean high reproducibility. Nat Mach Intell 2, 13–15 (2020). https://doi.org/10.1038/s42256-019-0140-2

Download citation

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1038/s42256-019-0140-2

This article is cited by

Search

Advanced search

Quick links

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing