Measuring feature importance of symbolic regression models using partial effects

In explainable AI, one aspect of a prediction's explanation is to measure each predictor's importance to the decision process. The importance can measure how much variation a predictor promotes locally or how much the predictor contributes to the deviation from a reference point (Shapley value). If we have the ground truth analytical model, we can calculate the former using the Partial Effect, calculated as the predictor's partial derivative. Also, we can estimate the latter by calculating the average partial effect multiplied by the difference between the predictor and the reference value. Symbolic Regression is a gray-box model for regression problems that returns an analytical model approximating the input data. Although it is often associated with interpretability, few works explore this property. This paper will investigate the use of Partial Effect with the analytical models generated by the Interaction-Transformation Evolutionary Algorithm symbolic regressor (ITEA). We show that the regression models returned by ITEA coupled with Partial Effect provide the closest explanations to the ground-truth and a close approximation to Shapley values. These results open up new opportunities to explain symbolic regression models compared to the approximations provided by model-agnostic approaches.

[1]  Dumitru Erhan,et al.  A Benchmark for Interpretability Methods in Deep Neural Networks , 2018, NeurIPS.

[2]  Scott Lundberg,et al.  A Unified Approach to Interpreting Model Predictions , 2017, NIPS.

[3]  Johannes Gehrke,et al.  Intelligible Models for HealthCare: Predicting Pneumonia Risk and Hospital 30-day Readmission , 2015, KDD.

[4]  Jason H. Moore,et al.  Where are we now?: a large benchmark study of recent symbolic regression methods , 2018, GECCO.

[5]  Carlos Guestrin,et al.  "Why Should I Trust You?": Explaining the Predictions of Any Classifier , 2016, ArXiv.

[6]  Thomas Lukasiewicz,et al.  Can I Trust the Explainer? Verifying Post-hoc Explanatory Methods , 2019, ArXiv.

[7]  Gabriel Kronberger,et al.  Parameter identification for symbolic regression using nonlinear least squares , 2019, Genetic Programming and Evolvable Machines.

[8]  Arthur K. Kordon,et al.  Prime-Time: Symbolic Regression Takes Its Place in the Real World , 2016 .

[9]  Marie-Jeanne Lesot,et al.  The Dangers of Post-hoc Interpretability: Unjustified Counterfactual Explanations , 2019, IJCAI.

[10]  Franco Turini,et al.  A Survey of Methods for Explaining Black Box Models , 2018, ACM Comput. Surv..

[11]  Christopher H. Achen.  Interpreting and Using Regression , 1982 .

[12]  Max Tegmark,et al.  AI Feynman: A physics-inspired method for symbolic regression , 2019, Science Advances.

[13]  Trevor Hastie,et al.  Causal Interpretations of Black-Box Models , 2019, Journal of business & economic statistics : a publication of the American Statistical Association.

[14]  Claudia Tarantola,et al.  Simple ways to interpret effects in modeling ordinal categorical data , 2018 .

[15]  Eberechukwu Onukwugha,et al.  A Primer on Marginal Effects—Part I: Theory and Formulae , 2014, PharmacoEconomics.

[16]  Erik Strumbelj,et al.  Explaining prediction models and individual predictions with feature contributions , 2014, Knowledge and Information Systems.

[17]  Sherif Sakr,et al.  Interpretability in HealthCare A Comparative Study of Local Machine Learning Interpretability Techniques , 2019, 2019 IEEE 32nd International Symposium on Computer-Based Medical Systems (CBMS).

[18]  Thomas Lukasiewicz,et al.  The Struggles of Feature-Based Explanations: Shapley Values vs. Minimal Sufficient Subsets , 2020, ArXiv.

[19]  Tommi S. Jaakkola,et al.  On the Robustness of Interpretability Methods , 2018, ArXiv.

[20]  Cynthia Rudin,et al.  Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead , 2018, Nature Machine Intelligence.

[21]  Mukund Sundararajan,et al.  The many Shapley values for model explanation , 2019, ICML.

[22]  Fabrício Olivetti de França,et al.  A greedy search tree heuristic for symbolic regression , 2018, Inf. Sci..

[23]  Stephan M. Winkler,et al.  Gaining Deeper Insights in Symbolic Regression , 2013, GPTP.

[24]  Guilherme Seidyo Imai Aldeia,et al.  A Parametric Study of Interaction-Transformation Evolutionary Algorithm for Symbolic Regression , 2020, 2020 IEEE Congress on Evolutionary Computation (CEC).

[25]  E. Norton,et al.  Marginal Effects-Quantifying the Effect of Changes in Risk Factors in Logistic Regression Models. , 2019, JAMA.

[26]  J. S. Long,et al.  A General Framework for Comparing Predictions and Marginal Effects across Models , 2019, Sociological Methodology.

[27]  Guilherme Seidyo Imai Aldeia,et al.  Lightweight Symbolic Regression with the Interaction - Transformation Representation , 2018, 2018 IEEE Congress on Evolutionary Computation (CEC).

[28]  Guilherme Seidyo Imai Aldeia,et al.  Interaction–Transformation Evolutionary Algorithm for Symbolic Regression , 2019, Evolutionary Computation.

[29]  Ronald L. Iman.  Latin Hypercube Sampling , 2008 .

[30]  Renato Miranda,et al.  Explaining Symbolic Regression Predictions , 2020, 2020 IEEE Congress on Evolutionary Computation (CEC).

[31]  Fabricio Olivetti de Franca,et al.  Interaction-transformation symbolic regression with extreme learning machine , 2021, Neurocomputing.

[32]  Rodrigo Silva,et al.  Applying Genetic Programming to Improve Interpretability in Machine Learning Models , 2020, 2020 IEEE Congress on Evolutionary Computation (CEC).

[33]  Been Kim,et al.  BIM: Towards Quantitative Evaluation of Interpretability Methods with Ground Truth , 2019, ArXiv.