Can local explanation techniques explain linear additive models? (2023-09-19T00:00:00.000000Z)

TL;DR

It is shown that even though the explanations generated by these techniques are linear additives, they can fail to provide accurate explanations when explaining linear additive models.

Abstract

Local model-agnostic additive explanation techniques decompose the predicted output of a black-box model into additive feature importance scores. Questions have been raised about the accuracy of the produced local additive explanations. We investigate this by studying whether some of the most popular explanation techniques can accurately explain the decisions of linear additive models. We show that even though the explanations generated by these techniques are linear additives, they can fail to provide accurate explanations when explaining linear additive models. In the experiments, we measure the accuracy of additive explanations, as produced by, e.g., LIME and SHAP, along with the non-additive explanations of Local Permutation Importance (LPI) when explaining Linear and Logistic Regression and Gaussian naive Bayes models over 40 tabular datasets. We also investigate the degree to which different factors, such as the number of numerical or categorical or correlated features, the predictive performance of the black-box model, explanation sample size, similarity metric, and the pre-processing technique used on the dataset can directly affect the accuracy of local explanations.

Authors

Amir Hossein Akhavan Rahnama

1 papers

Judith Bütepage

1 papers

Pierre Geurts

1 papers

TL;DR

Abstract

Authors

References45 items

OpenXAI: Towards a Transparent Evaluation of Model Explanations

Interpretability in the medical field: A systematic mapping and review study

When Comparing to Ground Truth is Wrong: On Evaluating GNN Explanation Methods

Synthetic Benchmarks for Scientific Research in Explainable Machine Learning

To trust or not to trust an explanation: using LEAF to evaluate local linear XAI methods

Evaluating local explanation methods on ground truth

Evaluating XAI: A comparison of rule-based and example-based explanations

Robust and Stable Black Box Explanations

On quantitative aspects of model interpretability

In Pursuit of Interpretable, Fair and Accurate Machine Learning for Criminal Recidivism Prediction

Visualizing the Impact of Feature Attribution Baselines

Towards Better Understanding of Adaptive Gradient Algorithms in Generative Adversarial Nets

A study of data and label shift in the LIME framework

Evaluations and Methods for Explanation through Robustness Analysis

Smooth Grad-CAM++: An Enhanced Inference Level Visualization Technique for Deep Convolutional Neural Network Models

Benchmarking Attribution Methods with Relative Feature Importance

Do Not Trust Additive Explanations

Explaining individual predictions when features are dependent: More accurate approximations to Shapley values

Please Stop Explaining Black Box Models for High Stakes Decisions

Sanity Checks for Saliency Maps

Model Agnostic Supervised Local Explanations

A Benchmark for Interpretability Methods in Deep Neural Networks

On the Robustness of Interpretability Methods

Towards Robust Interpretability with Self-Explaining Neural Networks

Defining Locality for Surrogates in Post-hoc Interpretablity

Anchors: High-Precision Model-Agnostic Explanations

Visualizing the Feature Importance for Black Box Models

Manipulating and Measuring Model Interpretability

Interpretation of Neural Networks is Fragile

Methods for interpreting and understanding deep neural networks

A Unified Approach to Interpreting Model Predictions

Interpretable Explanations of Black Boxes by Meaningful Perturbation

Towards A Rigorous Science of Interpretable Machine Learning

Model-Agnostic Interpretability of Machine Learning

Not Just a Black Box: Learning Important Features Through Propagating Activation Differences

“Why Should I Trust You?”: Explaining the Predictions of Any Classifier

Evaluating the Visualization of What a Deep Neural Network Has Learned

Comprehensible classification models: a position paper

Visualizing and Understanding Convolutional Networks

Introductory Statistics

Statistical Comparisons of Classifiers over Multiple Data Sets

Scikit-learn

Based on these scores, we describe how to measure the explanation accuracy of local explanation techniques directly

We present a novel principled method to extract the local ground truth model-intrinsic importance scores from additive terms in linear additive models

additive explanations explain

Field of Study

Journal Information

Name

Page

Volume

Venue Information

Name

Type

URL

Alternate Names