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

2019/03/25 by Kjersti Aas, Martin Jullum, Aas, Kjersti +3 · 130 citations
Computer Science · Mathematics · #Applied mathematics #Bayesian Modeling and Causal Inference #Computer science #Discrete mathematics #Explainable Artificial Intelligence (XAI) #Feature (linguistics) #Game theory #Kernel (algebra) #Machine Learning and Data Classification #Mathematical economics #Mathematical optimization #Mathematics #Series (stratigraphy) #Shapley value #Simple (philosophy) #cs.LG #stat.ME #stat.ML

paper · pdf · open access · doi:10.1016/j.artint.2021.103502

published in Artificial Intelligence 298, 103502 (Elsevier BV)

arxiv created 2020/02/06 · arxiv updated 2020/02/07 · openalex publication_date 2021/03/31 · openalex created_date 2025/10/10 · openalex updated_date 2026/08/05

Abstract

Explaining complex or seemingly simple machine learning models is an important practical problem. We want to explain individual predictions from a complex machine learning model by learning simple, interpretable explanations. Shapley values is a game theoretic concept that can be used for this purpose. The Shapley value framework has a series of desirable theoretical properties, and can in principle handle any predictive model. Kernel SHAP is a computationally efficient approximation to Shapley values in higher dimensions. Like several other existing methods, this approach assumes that the features are independent, which may give very wrong explanations. This is the case even if a simple linear model is used for predictions. In this paper, we extend the Kernel SHAP method to handle dependent features. We provide several examples of linear and non-linear models with various degrees of feature dependence, where our method gives more accurate approximations to the true Shapley values. We also propose a method for aggregating individual Shapley values, such that the prediction can be explained by groups of dependent variables.

Citations

• A NEW MEASURE OF RANK CORRELATION
• Remarks on Some Nonparametric Estimates of a Density Function

Cited by

• Local Additive Feature Attribution: A Mathematical Taxonomy and Reporting Checklist
• Feature Attribution-Based Explainability Analysis of Deep Learning Models in Predictive Process Monitoring
• Beyond Mean Solar Wind Conditions: Turbulence‐Aware Forecasting of the AE Index
• Explainable AI in Healthcare: to Explain, to Predict, or to Describe?
• FastSHAP: Real-Time Shapley Value Estimation
• Measuring Variable Importance via Accumulated Local Effects
• Interaction Tensor SHAP
• Machine learning for smell: Ordinal odor strength prediction of molecular perfumery components
• True to the Model or True to the Data?
• Feature-Function Curvature Analysis: A Geometric Framework for Explaining Differentiable Models
• Asymmetric Shapley values: incorporating causal knowledge into model-agnostic explainability
• On the Tractability of SHAP Explanations
• Opportunities and Challenges in Explainable Artificial Intelligence (XAI): A Survey
• On Locality of Local Explanation Models
• From Verification to Causality-based Explications
• The many Shapley values for model explanation
• Who stands out? Defining, measuring, and uncovering the elements of unusualness
• Spanish-language text classification for environmental evidence synthesis using multilingual pre-trained models
• Mortality Models Ensemble via Shapley Value
• Explaining by Removing: A Unified Framework for Model Explanation
• Data-driven Sensor Placement for Predictive Applications: A Correlation-Assisted Attribution Framework (CAAF)
• Statistical Crime Linkage: Evaluating approaches within the Covenant for Using AI in Policing
• Explainability for fair machine learning
• Stable multivariate lesion symptom mapping
• Oral Cancer Diagnosis Using Histopathology Images: An Explainable Hybrid Transformer Framework
• Can We Faithfully Represent Masked States to Compute Shapley Values on a DNN?
• SurvLIME-Inf: A simplified modification of SurvLIME for explanation of machine learning survival models
• Towards Trustworthy Vital Sign Forecasting: Leveraging Uncertainty for Prediction Intervals
• Causal SHAP: Feature Attribution with Dependency Awareness through Causal Discovery
• Shapley Flow: A Graph-based Approach to Interpreting Model Predictions
• Understanding Global Feature Contributions With Additive Importance Measures
• Breaking New Ground in Software Defect Prediction: Introducing Practical and Actionable Metrics with Superior Predictive Power for Enhanced Decision-Making
• On Using the Shapley Value for Anomaly Localization: A Statistical Investigation
• Explaining Algorithmic Fairness Through Fairness-Aware Causal Path Decomposition
• Explaining black box decisions by Shapley cohort refinement
• SHAFF: Fast and consistent SHApley eFfect estimates via random Forests
• Feature relevance quantification in explainable AI: A causal problem
• ConfPO: Exploiting Policy Model Confidence for Critical Token Selection in Preference Optimization
• TaylorPODA: A Taylor Expansion-Based Method to Improve Post-Hoc Attributions for Opaque Models
• midr: Learning from Black-Box Models by Maximum Interpretation Decomposition
• Feature Removal Is a Unifying Principle for Model Explanation Methods
• Fast TreeSHAP: Accelerating SHAP Value Computation for Trees
• An Interpretable Machine Learning Approach in Predicting Inflation Using Payments System Data: A Case Study of Indonesia
• Decorrelated Variable Importance
• A Unified Framework for Provably Efficient Algorithms to Estimate Shapley Values
• The Shapley Value of coalition of variables provides better explanations
• Prediction via Shapley Value Regression
• Towards Ground Truth Explainability on Tabular Data
• Causal Explanations Over Time: Articulated Reasoning for Interactive Environments
• Deep Argumentative Explanations
• Shapley explainability on the data manifold
• Explainable Machine Learning for Oxygen Diffusion in Perovskites and Pyrochlores
• Explainable Multimodal Machine Learning Using Combined Environmental DNA and Biogeographic Features for Ecosystem Biomonitoring
• Overview and practical recommendations on using Shapley Values for identifying predictive biomarkers via CATE modeling
• Measuring Explainer Stability via Attribution Separability
• DNN2LR: Automatic Feature Crossing for Credit Scoring
• What's Wrong with Your Synthetic Tabular Data? Using Explainable AI to Evaluate Generative Models
• Explanations Go Linear: Post-hoc Explainability for Tabular Data with Interpretable Meta-Encoding
• Fast approximate estimation of conditional Shapley values when using a linear explainer
• A gradient boosting approach for optimal selection of bidding strategies in reservoir hydro
• Are We Merely Justifying Results ex Post Facto? Quantifying Explanatory Inversion in Post-Hoc Model Explanations
• On the influence of dependent features in classification problems: A game-theoretic perspective
• Noninvasive Estimation of Hydration Status in Athletes Using Wearable Sensors and a Data-Driven Approach Based on Orthostatic Changes. [europepmc]
• Interpretable machine learning for genomics. [europepmc]
• Association Between Perioperative Adverse Cardiac Events and Mortality During One-Year Follow-Up After Noncardiac Surgery. [europepmc]
• Causal connections between socioeconomic disparities and COVID-19 in the USA. [europepmc]
• Multi-omics disease module detection with an explainable Greedy Decision Forest. [europepmc]
• Automated differential diagnosis of dementia syndromes using FDG PET and machine learning. [europepmc]
• Interpretable Machine Learning Techniques in ECG-Based Heart Disease Classification: A Systematic Review. [europepmc]
• Prediction model for myocardial injury after non-cardiac surgery using machine learning. [europepmc]
• Machine learning-based prediction model for postoperative delirium in non-cardiac surgery. [europepmc]
• WindowSHAP: An efficient framework for explaining time-series classifiers based on Shapley values. [europepmc]
• Distinct brain morphometry patterns revealed by deep learning improve prediction of post-stroke aphasia severity. [europepmc]
• Machine learning models identify micronutrient intake as predictors of undiagnosed hypertension among rural community-dwelling older adults in Thailand: a cross-sectional study. [europepmc]
• XAI-Based Assessment of the AMURA Model for Detecting Amyloid-β and Tau Microstructural Signatures in Alzheimer's Disease. [europepmc]
• Inequalities in global residential cooling energy use to 2050. [europepmc]
• Interpretable Machine Learning-Based Influence Factor Identification for 3D Printing Process-Structure Linkages. [europepmc]
• Visualization obesity risk prediction system based on machine learning. [europepmc]
• Agricultural practices influence soil microbiome assembly and interactions at different depths identified by machine learning. [europepmc]
• Iron Deficiency: Global Trends and Projections from 1990 to 2050. [europepmc]
• Towards verifiable cancer digital twins: tissue level modeling protocol for precision medicine. [europepmc]
• Predictive Mortality and Gastric Cancer Risk Using Clinical and Socio-Economic Data: A Nationwide Multicenter Cohort Study. [europepmc]
• Predicting major adverse cardiac events in diabetes and chronic kidney disease: a machine learning study from the Silesia Diabetes-Heart Project. [europepmc]
• Machine Learning and Spatio Temporal Analysis for Assessing Ecological Impacts of the Billion Tree Afforestation Project. [europepmc]
• A Shift Toward Supercritical Brain Dynamics Predicts Alzheimer's Disease Progression. [europepmc]
• Development of a Machine Learning-Based Predictive Model for Postoperative Delirium in Older Adult Intensive Care Unit Patients: Retrospective Study. [europepmc]
• Locating the missing absorption enhancement due to multi‒core black carbon aerosols. [europepmc]

Related

• Explaining the Model and Feature Dependencies by Decomposition of the Shapley Value
• Explaining the data or explaining a model? Shapley values that uncover non-linear dependencies
• Causal Shapley Values: Exploiting Causal Knowledge to Explain Individual Predictions of Complex Models
• Explaining predictive models with mixed features using Shapley values and conditional inference trees
• RKHS-SHAP: Shapley Values for Kernel Methods