Visualizing the Effects of Predictor Variables in Black Box Supervised Learning Models

Abstract

Summary In many supervised learning applications, understanding and visualizing the effects of the predictor variables on the predicted response is of paramount importance. A shortcoming of black box supervised learning models (e.g. complex trees, neural networks, boosted trees, random forests, nearest neighbours, local kernel‐weighted methods and support vector regression) in this regard is their lack of interpretability or transparency. Partial dependence plots, which are the most popular approach for visualizing the effects of the predictors with black box supervised learning models, can produce erroneous results if the predictors are strongly correlated, because they require extrapolation of the response at predictor values that are far outside the multivariate envelope of the training data. As an alternative to partial dependence plots, we present a new visualization approach that we term accumulated local effects plots, which do not require this unreliable extrapolation with correlated predictors. Moreover, accumulated local effects plots are far less computationally expensive than partial dependence plots. We also provide an R package ALEPlot as supplementary material to implement our proposed method.

Models of spatiotemporal variation in rabbit abundance reveal management hot spots for an invasive species

“…S3). Evaluation of these variables with accumulated local effects (ALE; Apley ) plots (Appendix : Fig. S4) suggests that at sites with lower temperature seasonality and lower mean annual temperatures in the years ( n = 2) prior to the survey, there was an increased probability of high rabbit densities.…”

“…Using these corrected abundances in a machine learning framework (random forest models; RF) provided a highly computationally efficient method for identifying the spatiotemporal drivers of heterogeneity in rabbit abundance across their invasive range. Furthermore, the use of RF permitted the analysis of non‐linear relationships between, and within, predictor variables, that may have otherwise been missed (Breiman , Apley ).…”

Using a supervised machine learning algorithm for detecting faking good in a personality self‐report

“…All of them had a quasi‐linear effect on XGB‐I predictions as it can be readily seen in their Accumulated Local Effects plots (Appendix in Supporting Information). These plots describe how single predictors influence model predictions (Apley & Zhu, ).…”

A Responsible Machine Learning Workflow with Focus on Interpretable Models, Post-hoc Explanation, and Discrimination Testing

“…PD and ICE can be overlaid in the same plot to create a holistic global and local portrait of the predictions for some g and X j [23]. When PD(X j , g) and ICE(x j , g) curves diverge, such plots can also be indicative of modeled interactions in g or expose flaws in PD estimation, e.g., inaccuracy in the presence of strong interactions and correlations [23,30]. For details regarding the calculation of one-dimensional PD and ICE, see the software resources in Section 2.8 and Appendices B.1 and B.4.…”