• Corpus ID: 237416659

Relating the Partial Dependence Plot and Permutation Feature Importance to the Data Generating Process

  title={Relating the Partial Dependence Plot and Permutation Feature Importance to the Data Generating Process},
  author={Christoph Molnar and Timo Freiesleben and Gunnar Konig and Giuseppe Casalicchio and Marvin N. Wright and B. Bischl},
Scientists and practitioners increasingly rely on machine learning to model data and draw conclusions. Compared to statistical modeling approaches, machine learning makes fewer explicit assumptions about data structures, such as linearity. However, their model parameters usually cannot be easily related to the data generating process. To learn about the modeled relationships, partial dependence (PD) plots and permutation feature importance (PFI) are often used as interpretation methods. However… 

Scientific Inference With Interpretable Machine Learning: Analyzing Models to Learn About Real-World Phenomena

A phenomenon-centric approach to IML in science clarifies the opportunities and limitations of IML for inference; that conditional not marginal sampling is required; and, the conditions under which the authors can trust IML methods.

Do AutoML-Based QSAR Models Fulfill OECD Principles for Regulatory Assessment? A 5-HT1A Receptor Case

This manuscript presents the best-in-class predictive model for the serotonin 1A receptor affinity and its validation according to the Organization for Economic Co-operation and Development guidelines for regulatory purposes and is developed based on a database with close to 9500 molecules by using an automatic machine learning tool (AutoML).

Data-Driven Estimation of a Driving Safety Tolerance Zone Using Imbalanced Machine Learning

A framework to identify the level of risky driving behavior as well as the duration of the time spent in each risk level by private car drivers is proposed and the maximum speed of the vehicle during a 30 s interval, is the most crucial predictor for identifying the driving time at each safety level.

Exploring Associations between Multimodality and Built Environment Characteristics in the U.S

This study demonstrated associations between multimodality and built environment characteristics, and proposed policy implications for fostering multimodal travel behaviors. It conducted a U.S.

Satisfaction with the Pedestrian Environment and Its Relationship to Neighborhood Satisfaction in Seoul, South Korea

This study investigated the relationship between the degree of satisfaction with the pedestrian environments in their neighborhoods and the degree of neighborhood satisfaction in Seoul, South Korea.



Permutation importance: a corrected feature importance measure

A heuristic for normalizing feature importance measures that can correct the feature importance bias is introduced and PIMP was used to correct RF-based importance measures for two real-world case studies and improve model interpretability.

Understanding Variable Effects from Black Box Prediction: Quantifying Effects in Tree Ensembles Using Partial Dependence

A method for quantifying variable effects using partial dependence is described, which produces an estimate that can be interpreted as the effect on the response for a one unit change in the predictor, while averaging over the effects of all other variables.

Testing conditional independence in supervised learning algorithms

A novel testing procedure is developed that works in conjunction with any valid knockoff sampler, supervised learning algorithm, and loss function, and demonstrates convergence criteria for the CPI and develops statistical inference procedures for evaluating its magnitude, significance, and precision.

All Models are Wrong, but Many are Useful: Learning a Variable's Importance by Studying an Entire Class of Prediction Models Simultaneously

Model class reliance (MCR) is proposed as the range of VI values across all well-performing model in a prespecified class, which gives a more comprehensive description of importance by accounting for the fact that many prediction models, possibly of different parametric forms, may fit the data well.

Model-agnostic Feature Importance and Effects with Dependent Features - A Conditional Subgroup Approach

This work proposes conditional variants of partial dependence plots and permutation feature importance that reduce extrapolation and introduces a data fidelity measure that captures the degree of extrapolation when data is transformed with a certain perturbation.

Please Stop Permuting Features: An Explanation and Alternatives

This paper argues that breaking dependencies between features in hold-out data places undue emphasis on sparse regions of the feature space by forcing the original model to extrapolate to regions where there is little to no data, and finds support for previous claims in the literature that PaP metrics tend to over-emphasize correlated features.

A Stratification Approach to Partial Dependence for Codependent Variables

StratPD is introduced, a new strategy for assessing partial dependence that does not depend on a user’s fitted model, provides accurate results in the presence codependent variables, and is applicable to high dimensional settings.

Double/Debiased Machine Learning for Treatment and Structural Parameters

This work revisits the classic semiparametric problem of inference on a low dimensional parameter θ_0 in the presence of high-dimensional nuisance parameters η_0 and proves that DML delivers point estimators that concentrate in a N^(-1/2)-neighborhood of the true parameter values and are approximately unbiased and normally distributed, which allows construction of valid confidence statements.

Visualizing the effects of predictor variables in black box supervised learning models

  • D. ApleyJingyu Zhu
  • Computer Science
    Journal of the Royal Statistical Society: Series B (Statistical Methodology)
  • 2020
Accumulated local effects plots are presented, which do not require this unreliable extrapolation with correlated predictors and are far less computationally expensive than partial dependence plots.

Sampling, Intervention, Prediction, Aggregation: A Generalized Framework for Model Agnostic Interpretations

The generalized SIPA (Sampling, Intervention, Prediction, Aggregation) framework of work stages for model agnostic interpretation techniques is presented and several prominent methods for feature effects can be embedded into the proposed framework.