IDEAS home Printed from https://ideas.repec.org/p/arx/papers/2103.11706.html
   My bibliography  Save this paper

Interpreting Deep Learning Models with Marginal Attribution by Conditioning on Quantiles

Author

Listed:
  • M. Merz
  • R. Richman
  • T. Tsanakas
  • M. V. Wuthrich

Abstract

A vastly growing literature on explaining deep learning models has emerged. This paper contributes to that literature by introducing a global gradient-based model-agnostic method, which we call Marginal Attribution by Conditioning on Quantiles (MACQ). Our approach is based on analyzing the marginal attribution of predictions (outputs) to individual features (inputs). Specificalllly, we consider variable importance by mixing (global) output levels and, thus, explain how features marginally contribute across different regions of the prediction space. Hence, MACQ can be seen as a marginal attribution counterpart to approaches such as accumulated local effects (ALE), which study the sensitivities of outputs by perturbing inputs. Furthermore, MACQ allows us to separate marginal attribution of individual features from interaction effect, and visually illustrate the 3-way relationship between marginal attribution, output level, and feature value.

Suggested Citation

  • M. Merz & R. Richman & T. Tsanakas & M. V. Wuthrich, 2021. "Interpreting Deep Learning Models with Marginal Attribution by Conditioning on Quantiles," Papers 2103.11706, arXiv.org.
  • Handle: RePEc:arx:papers:2103.11706
    as

    Download full text from publisher

    File URL: http://arxiv.org/pdf/2103.11706
    File Function: Latest version
    Download Restriction: no
    ---><---

    References listed on IDEAS

    as
    1. Guangwu Liu & Liu Jeff Hong, 2009. "Kernel estimation of quantile sensitivities," Naval Research Logistics (NRL), John Wiley & Sons, vol. 56(6), pages 511-525, September.
    2. Michael C. Fu & L. Jeff Hong & Jian-Qiang Hu, 2009. "Conditional Monte Carlo Estimation of Quantile Sensitivities," Management Science, INFORMS, vol. 55(12), pages 2019-2027, December.
    3. Bradley Efron, 2020. "Prediction, Estimation, and Attribution," Journal of the American Statistical Association, Taylor & Francis Journals, vol. 115(530), pages 636-655, April.
    4. Daniel W. Apley & Jingyu Zhu, 2020. "Visualizing the effects of predictor variables in black box supervised learning models," Journal of the Royal Statistical Society Series B, Royal Statistical Society, vol. 82(4), pages 1059-1086, September.
    5. Bradley Efron, 2020. "Prediction, Estimation, and Attribution," International Statistical Review, International Statistical Institute, vol. 88(S1), pages 28-59, December.
    6. Wang, Shaun, 1996. "Premium Calculation by Transforming the Layer Premium Density," ASTIN Bulletin, Cambridge University Press, vol. 26(1), pages 71-92, May.
    7. L. Jeff Hong, 2009. "Estimating Quantile Sensitivities," Operations Research, INFORMS, vol. 57(1), pages 118-130, February.
    8. Qingyuan Zhao & Trevor Hastie, 2021. "Causal Interpretations of Black-Box Models," Journal of Business & Economic Statistics, Taylor & Francis Journals, vol. 39(1), pages 272-281, January.
    Full references (including those not matched with items on IDEAS)

    Citations

    Citations are extracted by the CitEc Project, subscribe to its RSS feed for this item.
    as


    Cited by:

    1. Ronald Richman & Mario V. Wuthrich, 2021. "LocalGLMnet: interpretable deep learning for tabular data," Papers 2107.11059, arXiv.org.

    Most related items

    These are the items that most often cite the same works as this one and are cited by the same works as this one.
    1. Makam, Vaishno Devi & Millossovich, Pietro & Tsanakas, Andreas, 2021. "Sensitivity analysis with χ2-divergences," Insurance: Mathematics and Economics, Elsevier, vol. 100(C), pages 372-383.
    2. Pesenti, Silvana M. & Tsanakas, Andreas & Millossovich, Pietro, 2018. "Euler allocations in the presence of non-linear reinsurance: Comment on Major (2018)," Insurance: Mathematics and Economics, Elsevier, vol. 83(C), pages 29-31.
    3. Silvana M. Pesenti & Pietro Millossovich & Andreas Tsanakas, 2023. "Differential Sensitivity in Discontinuous Models," Papers 2310.06151, arXiv.org.
    4. Xi Chen & Kyoung-Kuk Kim, 2016. "Efficient VaR and CVaR Measurement via Stochastic Kriging," INFORMS Journal on Computing, INFORMS, vol. 28(4), pages 629-644, November.
    5. Pesenti, Silvana M. & Millossovich, Pietro & Tsanakas, Andreas, 2019. "Reverse sensitivity testing: What does it take to break the model?," European Journal of Operational Research, Elsevier, vol. 274(2), pages 654-670.
    6. Jiaqiao Hu & Yijie Peng & Gongbo Zhang & Qi Zhang, 2022. "A Stochastic Approximation Method for Simulation-Based Quantile Optimization," INFORMS Journal on Computing, INFORMS, vol. 34(6), pages 2889-2907, November.
    7. Katja Schilling & Daniel Bauer & Marcus C. Christiansen & Alexander Kling, 2020. "Decomposing Dynamic Risks into Risk Components," Management Science, INFORMS, vol. 66(12), pages 5738-5756, December.
    8. J P C Kleijnen & W C M van Beers, 2013. "Monotonicity-preserving bootstrapped Kriging metamodels for expensive simulations," Journal of the Operational Research Society, Palgrave Macmillan;The OR Society, vol. 64(5), pages 708-717, May.
    9. Weihuan Huang, 2023. "Estimating Systemic Risk within Financial Networks: A Two-Step Nonparametric Method," Papers 2310.18658, arXiv.org.
    10. Yijie Peng & Michael C. Fu & Bernd Heidergott & Henry Lam, 2020. "Maximum Likelihood Estimation by Monte Carlo Simulation: Toward Data-Driven Stochastic Modeling," Operations Research, INFORMS, vol. 68(6), pages 1896-1912, November.
    11. Guangxin Jiang & Michael C. Fu, 2015. "Technical Note—On Estimating Quantile Sensitivities via Infinitesimal Perturbation Analysis," Operations Research, INFORMS, vol. 63(2), pages 435-441, April.
    12. Peter W. Glynn & Yijie Peng & Michael C. Fu & Jian-Qiang Hu, 2021. "Computing Sensitivities for Distortion Risk Measures," INFORMS Journal on Computing, INFORMS, vol. 33(4), pages 1520-1532, October.
    13. Weihuan Huang & Nifei Lin & L. Jeff Hong, 2022. "Monte-Carlo Estimation of CoVaR," Papers 2210.06148, arXiv.org.
    14. Bernd Heidergott & Warren Volk-Makarewicz, 2016. "A Measure-Valued Differentiation Approach to Sensitivities of Quantiles," Mathematics of Operations Research, INFORMS, vol. 41(1), pages 293-317, February.
    15. Andreas Tsanakas & Pietro Millossovich, 2016. "Sensitivity Analysis Using Risk Measures," Risk Analysis, John Wiley & Sons, vol. 36(1), pages 30-48, January.
    16. L. Jeff Hong & Yi Yang & Liwei Zhang, 2011. "Sequential Convex Approximations to Joint Chance Constrained Programs: A Monte Carlo Approach," Operations Research, INFORMS, vol. 59(3), pages 617-630, June.
    17. Guangwu Liu, 2015. "Simulating Risk Contributions of Credit Portfolios," Operations Research, INFORMS, vol. 63(1), pages 104-121, February.
    18. He, Zhijian, 2022. "Sensitivity estimation of conditional value at risk using randomized quasi-Monte Carlo," European Journal of Operational Research, Elsevier, vol. 298(1), pages 229-242.
    19. Koike, Takaaki & Saporito, Yuri & Targino, Rodrigo, 2022. "Avoiding zero probability events when computing Value at Risk contributions," Insurance: Mathematics and Economics, Elsevier, vol. 106(C), pages 173-192.
    20. Yongqiang Wang & Michael C. Fu & Steven I. Marcus, 2012. "A New Stochastic Derivative Estimator for Discontinuous Payoff Functions with Application to Financial Derivatives," Operations Research, INFORMS, vol. 60(2), pages 447-460, April.

    More about this item

    Statistics

    Access and download statistics

    Corrections

    All material on this site has been provided by the respective publishers and authors. You can help correct errors and omissions. When requesting a correction, please mention this item's handle: RePEc:arx:papers:2103.11706. See general information about how to correct material in RePEc.

    If you have authored this item and are not yet registered with RePEc, we encourage you to do it here. This allows to link your profile to this item. It also allows you to accept potential citations to this item that we are uncertain about.

    If CitEc recognized a bibliographic reference but did not link an item in RePEc to it, you can help with this form .

    If you know of missing items citing this one, you can help us creating those links by adding the relevant references in the same way as above, for each refering item. If you are a registered author of this item, you may also want to check the "citations" tab in your RePEc Author Service profile, as there may be some citations waiting for confirmation.

    For technical questions regarding this item, or to correct its authors, title, abstract, bibliographic or download information, contact: arXiv administrators (email available below). General contact details of provider: http://arxiv.org/ .

    Please note that corrections may take a couple of weeks to filter through the various RePEc services.

    IDEAS is a RePEc service. RePEc uses bibliographic data supplied by the respective publishers.