An official website of the United States government
Abstract With increasing interest in explaining machine learning (ML) models, this paper synthesizes many topics related to ML explainability. We distinguish explainability from interpretability, local from global explainability, and feature importance versus feature relevance. We demonstrate and visualize different explanation methods, how to interpret them, and provide a complete Python package (scikit-explain) to allow future researchers and model developers to explore these explainability methods. The explainability methods include Shapley additive explanations (SHAP), Shapley additive global explanation (SAGE), and accumulated local effects (ALE). Our focus is primarily on Shapley-based techniques, which serve as a unifying framework for various existing methods to enhance model explainability. For example, SHAP unifies methods like local interpretable model-agnostic explanations (LIME) and tree interpreter for local explainability, while SAGE unifies the different variations of permutation importance for global explainability. We provide a short tutorial for explaining ML models using three disparate datasets: a convection-allowing model dataset for severe weather prediction, a nowcasting dataset for subfreezing road surface prediction, and satellite-based data for lightning prediction. In addition, we showcase the adverse effects that correlated features can have on the explainability of a model. Finally, we demonstrate the notion of evaluating model impacts of feature groups instead of individual features. Evaluating the feature groups mitigates the impacts of feature correlations and can provide a more holistic understanding of the model. All code, models, and data used in this study are freely available to accelerate the adoption of machine learning explainability in the atmospheric and other environmental sciences.
more »
« less
Marginal Contribution Feature Importance - an Axiomatic Approach for Explaining Data
In recent years, methods were proposed for assigning feature importance scores to measure the contribution of individual features. While in some cases the goal is to understand a specific model, in many cases the goal is to understand the contribution of certain properties (features) to a real-world phenomenon. Thus, a distinction has been made between feature importance scores that explain a model and scores that explain the data. When explaining the data, machine learning models are used as proxies in settings where conducting many real-world experiments is expensive or prohibited. While existing feature importance scores show great success in explaining models, we demonstrate their limitations when explaining the data, especially in the presence of correlations between features. Therefore, we develop a set of axioms to capture properties expected from a feature importance score when explaining data and prove that there exists only one score that satisfies all of them, the Marginal Contribution Feature Importance (MCI). We analyze the theoretical properties of this score function and demonstrate its merits empirically.
more »
« less
- Award ID(s):
- 1705121
- PAR ID:
- 10366172
- Date Published:
- Journal Name:
- Proceedings of the 38th International Conference on Machine Learning
- Volume:
- 139
- Page Range / eLocation ID:
- 1324-1335
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Explaining machine learning models is an important and increasingly popular area of research interest. The Shapley value from game theory has been proposed as a prime approach to compute feature importance towards model predictions on images, text, tabular data, and recently graph neural networks (GNNs) on graphs. In this work, we revisit the appropriateness of the Shapley value for GNN explanation, where the task is to identify the most important subgraph and constituent nodes for GNN predictions. We claim that the Shapley value is a non-ideal choice for graph data because it is by definition not structure-aware. We propose a Graph Structure-aware eXplanation (GStarX) method to leverage the critical graph structure information to improve the explanation. Specifically, we define a scoring function based on a new structure-aware value from the cooperative game theory proposed by Hamiache and Navarro (HN). When used to score node importance, the HN value utilizes graph structures to attribute cooperation surplus between neighbor nodes, resembling message passing in GNNs, so that node importance scores reflect not only the node feature importance, but also the node structural roles. We demonstrate that GStarX produces qualitatively more intuitive explanations, and quantitatively improves explanation fidelity over strong baselines on chemical graph property prediction and text graph sentiment classification.more » « less
-
Feature importance is commonly used to explain machine predictions. While feature importance can be derived from a machine learning model with a variety of methods, the consistency of feature importance via different methods remains understudied. In this work, we systematically compare feature importance from built-in mechanisms in a model such as attention values and post-hoc methods that approximate model behavior such as LIME. Using text classification as a testbed, we find that 1) no matter which method we use, important features from traditional models such as SVM and XGBoost are more similar with each other, than with deep learning models; 2) post-hoc methods tend to generate more similar important features for two models than built-in methods. We further demonstrate how such similarity varies across instances. Notably, important features do not always resemble each other better when two models agree on the predicted label than when they disagree.more » « less
-
In explaining how humans selectively attend, common frameworks often focus on how attention is allocated relative to an idealized allocation based on properties of the task. However, these perspectives often ignore different types of constraints that could help explain why attention was allocated in a particular way. For example, many computational models of learning are well equipped to explain how attention should ideally be allocated to minimize errors within the task, but these models often assume all features are perfectly encoded or that the only learning goal is to maximize accuracy. In this article, we argue for a more comprehensive view by using computational modeling to understand the complex interactions that occur between selective attention and memory. Our central thesis is that although selective attention directs attention to relevant dimensions, relevance can be established only through memories of previous experiences. Hence, attention is initially used to encode features and create memories, but thereafter, attention operates selectively on the basis of what is kept in memory. Through this lens, deviations from ideal performance can still be viewed as goal-directed selective attention, but the orientation of attention is subject to the constraints of the individual learner.more » « less
-
Operational networks commonly rely on machine learning models for many tasks, including detecting anomalies, inferring application performance, and forecasting demand. Yet, model accuracy can degrade due to concept drift, whereby the relationship between the features and the target to be predicted changes. Mitigating concept drift is an essential part of operationalizing machine learning models in general, but is of particular importance in networking's highly dynamic deployment environments. In this paper, we first characterize concept drift in a large cellular network for a major metropolitan area in the United States. We find that concept drift occurs across many important key performance indicators (KPIs), independently of the model, training set size, and time interval---thus necessitating practical approaches to detect, explain, and mitigate it. We then show that frequent model retraining with newly available data is not sufficient to mitigate concept drift, and can even degrade model accuracy further. Finally, we develop a new methodology for concept drift mitigation, Local Error Approximation of Features (LEAF). LEAF works by detecting drift; explaining the features and time intervals that contribute the most to drift; and mitigates it using forgetting and over-sampling. We evaluate LEAF against industry-standard mitigation approaches (notably, periodic retraining) with more than four years of cellular KPI data. Our initial tests with a major cellular provider in the US show that LEAF consistently outperforms periodic and triggered retraining on complex, real-world data while reducing costly retraining operations.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- The DOI is not currently available.
