A Unified Approach to Interpreting Model Predictions
Единый подход к интерпретации предсказаний моделей
2017-05-22
SCID: 54.1/77tkzhrx
Discuss with AI
SHAPShapley additive explanationsadditive feature attributionfeature importancemodel interpretability
Figures from the paper
Abstract (AI)
Understanding why a model makes a certain prediction can be as crucial as the prediction's accuracy in many applications. However, the highest accuracy for large modern datasets is often achieved by complex models that even experts struggle to interpret, such as ensemble or deep learning models, creating a tension between accuracy and interpretability. In response, various methods have recently been proposed to help users interpret the predictions of complex models, but it is often unclear how these methods are related and when one method is preferable over another. To address this problem, we present a unified framework for interpreting predictions, SHAP (SHapley Additive exPlanations). SHAP assigns each feature an importance value for a particular prediction. Its novel components include: (1) the identification of a new class of additive feature importance measures, and (2) theoretical results showing there is a unique solution in this class with a set of desirable properties. The new class unifies six existing methods, notable because several recent methods in the class lack the proposed desirable properties. Based on insights from this unification, we present new methods that show improved computational performance and/or better consistency with human intuition than previous approaches.
Key Findings
1
Insights from the unification yield new methods with improved computational performance and/or greater consistency with human intuition.
2
SHAP provides a unified framework for interpreting individual predictions by assigning each feature an importance value.
3
The framework identifies a new class of additive feature-importance measures for explaining model predictions.
4
The framework unifies six existing prediction-interpretation methods, several of which lack the proposed desirable properties.
5
Within this class, SHAP is theoretically shown to be the unique solution satisfying a specified set of desirable properties.
Research Object
Predictions of complex machine-learning models
Research Subject
Feature-importance attribution for individual predictions, including the unification, axiomatic characterization, and computational improvement of additive explanation methods
Publication Details
Publication Date
2017-05-22
Journal
Publisher
ISSN
Cited by
54521
Open access PDF
Access Type
Author Information
Download PDF
Subscribe to digest
References available in scid.ai1
Cited by20
Peeking Inside the Black-Box: A Survey on Explainable Artificial Intelligence (XAI)2018
Interpreting Black-Box Models: A Review on Explainable Artificial Intelligence2023
A Survey on Explainable Artificial Intelligence (XAI): Toward Medical XAI2020
CatBoost for big data: an interdisciplinary review2020
Time-series forecasting with deep learning: a survey2021
Explainable Artificial Intelligence (XAI): What we know and what is left to attain Trustworthy Artificial Intelligence2023
Machine Learning Interpretability: A Survey on Methods and Metrics2019
Machine Learning and Deep Learning Methods for Intrusion Detection Systems: A Survey2019
Deep learning in cancer diagnosis, prognosis and treatment selection2021
A Survey on the Explainability of Supervised Machine Learning2021
Fairness in Machine Learning: A Survey2023
Questioning the AI: Informing Design Practices for Explainable AI User Experiences2020
Designing Theory-Driven User-Centric Explainable AI2019
From What to How: An Initial Review of Publicly Available AI Ethics Tools, Methods and Research to Translate Principles into Practices2019
Bias in medical AI: Implications for clinical decision-making2024
Financial Fraud: A Review of Anomaly Detection Techniques and Recent Advances2021
Self-Driving Laboratories for Chemistry and Materials Science2024
Industrial Artificial Intelligence in Industry 4.0 - Systematic Review, Challenges and Outlook2020
Explainability for Large Language Models: A Survey2024
Explainable artificial intelligence (XAI) in finance: a systematic literature review2024