Abstract
Machine learning models have been widely used in various domains. However, the internal mechanisms of popular models, such as neural networks and support vector machines, are difficult for humans to understand; such models are often called “black boxes”. In this study, a general method is proposed to gain insight into the black boxes of supervised learning models by visualizing the impacts of input features on their prediction results. Compared with the existing methods, which may overlook the overall understanding of prediction models by analyzing the feature impacts for each individual observation or ignore the impact differences by providing a single impact pattern for all observations, the proposed method distinguishes some typical impact patterns that correspond to different groups of observations. The method maps the detected impact patterns into feature space using tree rules that help locate the impact patterns in the feature space. More importantly, the feature relationships embedded in the prediction models can be revealed through this tree rule-based feature relationship network. We apply the proposed method to various simulated and real data, and the results demonstrate how it can help us understand how features affect model prediction results and the relationships among features.
Similar content being viewed by others
Code Availability
the codes developed in the research are available at https://github.com/XZPackage/IPPModel.
References
Ancona M, Ceolini E, Öztireli C, Gross M (2018) Towards better understanding of gradient-based attribution methods for deep neural networks. In: Proceedings of the international conference on learning representation
Breiman L, Cutler A, Liaw A, Wiener M (2015) Package ‘randomforest’. https://cran.r-project.org/web/packages/randomForest/randomForest.pdf
Breiman L, Friedman JH, Olshen RA, Stone CJ (1984) Classification and regression trees. Wadsworth, Belmont
Carrington AM, Fieguth PW, et al. (2020) A new concordant partial AUC and partial c statistic for imbalanced data in the evaluation of machine learning algorithms. BMC Med Inform Decis Mak 20 (4):1–12
Charrad M, Ghazzali N, Boiteau V, Niknafs A (2014) Nbclust: An R package for determining the relevant number of clusters in a data set. J Stat Softw 061(i06):1–36
Cortez P, Cerdeira A, Almeida F, Matos T, Reis J (2009) Modeling wine preferences by data mining from physicochemical properties. Decis Support Syst 47(4):547–553
Cortez P, Embrechts MJ (2013) Using sensitivity analysis and visualization techniques to open black box data mining models. Inf Sci 225:1–17
Doron M, Segev I, Shahaf D (2019) Discovering unexpected local nonlinear interactions in scientific black-box models. In: Proceedings of the 25th ACM SIGKDD international conference on knowledge discovery and data mining, pp 425–43
Dunn JC (1974) Well-separated clusters and optimal fuzzy partitions. J Cybern 4:95–104
Friedman JH (2001) Greedy function approximation: A gradient boosting machine. Ann Stat 29 (5):1189–1232
Goldstein A, Kapelner A, Bleich J, Pitkin E (2015) Peeking inside the black box: Visualizing statistical learning with plots of individual conditional expectation. J Comput Graph Stat 24(1):44–65
Guidotti R, Monreale A, Ruggieri S, Turini F, Giannotti F, Pedreschi D (2018) A survey of methods for explaining black box models. ACM Comput Surv 51(5):1–42
Guidotti R, Monreale A, Giannotti F, Pedreschi D, Ruggieri S, Turini F (2019) Factual and counterfactual explanations for black box decision making. IEEE Intell Syst 34(6):14–23
Henelius A, Puolamaki K, Bostrom H, Asker L, Papapetrou P (2014) A peek into the black box: Exploring classifers by randomization. Data Min Knowl Discov 9:1503–1529
Henelius A, Puolamaki K, Karlsson I, Zhao J, Asker L, Bostrom H, Papapetrou P (2015) Goldeneye++: A closer look into the black box. In: Proceedings of statistical learning and data sciences: Third international symposium, pp 96–105
Hooker G (2004) Discovering additive structure in black box functions. In: Proceedings of the Tenth ACM SIGKDD international conference on knowledge discovery and data mining , pp 575–580
Karatzoglou A, Smola A, Hornik K, Zeileis A (2004) Kernlab - an S4 package for kernel methods in R. J Stat Softw 1(9):1–20
Krause J, Perer A, Bertini E (2016) Using visual analytics to interpret predictive machine learning models. In: ICML workshop on human interpretability in machine learning, pp 1–5
Li X-H, Gao CC, Shi Y, Bai W, Gao H, Qiu L, Wang C, Gao Y, Zhang S, Xue X et al (2020) A survey of data-driven and knowledge-aware explainable AI. IEEE Trans Knowl Data Eng https://doi.org/10.1109/TKDE.2020.2983930
Lundberg SM, Lee S-I (2017) A unified approach to interpreting model predictions. In: Advances in neural information processing systems, vol 30, pp 4765–4774
Moro S, Cortez P, Rita P (2014) A data-driven approach to predict the success of bank telemarketing. Decis Support Syst 62:22–31
Oh S (2019) Feature interaction in terms of prediction performance. Appl Sci 9(23):5191
Ojala M, Garriga GC (2010) Permutation tests for studying classifier performance. J Mach Learn Res 11:1833–1633
Ribeiro MT, Singh S, Guestrin C (2016) Why should I trust you?: Explaining the predictions of any classifier. In: Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining, pp 1135–1144
Ripley B, Venables W (2016) Package ‘nnet’. https://cran.r-project.org/web/packages/nnet/nnet.pdf
Sundararajan M, Taly A, Yan Q (2017) Axiomatic attribution for deep networks. In: Proceedings of the 34th international conference on machine learning, vol 70, pp 3319–3328
Tibshirani R, Walther G, Hastie T (2001) Estimating the number of clusters in a data set via the gap statistic. J R Stat Soc Ser B-Stat Methodol 63(2):411–423
Acknowledgments
This study was partially supported by National Key R&D Program of China (2018YFC0830801).
Funding
This study was partially supported by National Key R&D Program of China (2018YFC0830801).
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of Interests
no conflicts of interest and no computing interests.
Additional information
Availability of Data and Material
the data used in the research is generated randomly or available publicly.
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Electronic supplementary material
Below is the link to the electronic supplementary material.
Rights and permissions
About this article
Cite this article
Zhang, X., Wang, Y. & Li, Z. Interpreting the black box of supervised learning models: Visualizing the impacts of features on prediction. Appl Intell 51, 7151–7165 (2021). https://doi.org/10.1007/s10489-021-02255-z
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10489-021-02255-z