Conditional Feature Importance with Generative Modeling Using Adversarial Random Forests

Abstract

This paper proposes a method for measuring conditional feature importance via generative modeling. In explainable artificial intelligence (XAI), conditional feature importance assesses the impact of a feature on a prediction model's performance given the information of other features. Model-agnostic post hoc methods to do so typically evaluate changes in the predictive performance under on-manifold feature value manipulations. Such procedures require creating feature values that respect conditional feature distributions, which can be challenging in practice. Recent advancements in generative modeling can facilitate this. For tabular data, which may consist of both categorical and continuous features, the adversarial random forest (ARF) stands out as a generative model that can generate on-manifold data points without requiring intensive tuning efforts or computational resources, making it a promising candidate model for subroutines in XAI methods. This paper proposes cARFi (conditional ARF feature importance), a method for measuring conditional feature importance through feature values sampled from ARF-estimated conditional distributions. cARFi requires only little tuning to yield robust importance scores that can flexibly adapt for conditional or marginal notions of feature importance, including straightforward extensions to condition on feature subsets and allows for inferring the significance of feature importances through statistical tests.

Figures (12)

• Figure 1: Rejection rates of one-sided paired $t$-tests at $\alpha ~= 0.05$ to detect relevant features at different effect sizes, i.e., type I error rates at effect size 0 and power at effect size $>$0.
• Figure 2: Rejection rates of one-sided paired $t$-tests at $\alpha ~= 0.05$ to detect relevant features, i.e. power and type I error rates, across $500$ simulation runs. $X_1, X_3$ are 10-level categoricals, $X_2, X_4$ are Gaussian. Effect size $\beta ~= 0.5$ and random forest prediction model.
• Figure 3: Marginal vs. conditional feature importances for a linear model (LM, upper row) and a random forest (RF, lower row). While the LM coefficients $\beta_1, \beta_2, \beta_5$ are close to zero, the RF assigns marginal importance to $X_1, X_2, X_5$ due to their strong correlation with $X_3, X_4$. cARFi resolves these indirect influences by conditioning on the respective feature subsets.
• Figure 4: Feature importance values for the variables from the real-world bike rental example. Panel A: PFI and cARFi (conditioned on all other ones) values of all included variables in the random forest model. Panels B and C: cARFi values for Hour and Temperature for selected conditioning sets, respectively. The RMSE is used as the loss function and 50 repetitions.
• Figure S5: Proof of concept for minimum leaf size of 2.