Skip to main content
QUICK REVIEW

[论文解读] Grouped Feature Importance and Combined Features Effect Plot

Quay Au, Julia Herbinger|arXiv (Cornell University)|Apr 23, 2021
Explainable Artificial Intelligence (XAI)被引用 10
一句话总结

本文提出了分组特征重要性(GFI)与组合特征效应图(CFEP),通过置换、重拟合及基于Shapley的方法,将模型无关的可解释性扩展至特征组。提出一种顺序程序以识别稳定且高性能的特征组组合,并通过模拟实验与心理学数据集验证,GFI与CFEP在相关性或高维数据中显著提升可解释性,尤其在单特征分析具有误导性时表现更优。

ABSTRACT

Interpretable machine learning has become a very active area of research due to the rising popularity of machine learning algorithms and their inherently challenging interpretability. Most work in this area has been focused on the interpretation of single features in a model. However, for researchers and practitioners, it is often equally important to quantify the importance or visualize the effect of feature groups. To address this research gap, we provide a comprehensive overview of how existing model-agnostic techniques can be defined for feature groups to assess the grouped feature importance, focusing on permutation-based, refitting, and Shapley-based methods. We also introduce an importance-based sequential procedure that identifies a stable and well-performing combination of features in the grouped feature space. Furthermore, we introduce the combined features effect plot, which is a technique to visualize the effect of a group of features based on a sparse, interpretable linear combination of features. We used simulation studies and a real data example from computational psychology to analyze, compare, and discuss these methods.

研究动机与目标

  • 解决机器学习模型中特征组可解释性缺乏模型无关方法的问题。
  • 将现有单特征可解释性技术(如置换重要性与Shapley值)扩展至特征组。
  • 开发一种顺序程序,以识别稳定且高性能的特征组组合。
  • 提出组合特征效应图(CFEP),用于可视化特征组对模型预测的联合效应。
  • 在不同相关结构与数据类型下,评估分组重要性方法的性能与可靠性。

提出的方法

  • 提出两种分组特征重要性的策略:从零模型或全模型出发,分别适用于置换与重拟合方法。
  • 引入基于置换与重拟合方法的分组特征Shapley重要性评分。
  • 开发一种顺序分组特征重要性程序,通过迭代选择基于稳定性和预测性能的特征组。
  • 提出组合特征效应图(CFEP),利用监督稀疏主成分分析(SPCA)生成可解释的特征线性组合以实现可视化。
  • 通过SPCA进行降维,以确保CFEP中的可解释性与稀疏性,优先采用监督SPCA以提升性能。
  • 通过模拟研究与真实世界计算心理学数据集,对比并验证所提出方法。
Figure 1 : Sequential grouped feature importance for the simulation in Section 3.1 . 100 times repeated subsampling. Improvement threshold $\delta=0.001$ . Vertical bars show one step of the sequential procedure (left to right). Height of the vertical bars represent the number of subsampling iterati
Figure 1 : Sequential grouped feature importance for the simulation in Section 3.1 . 100 times repeated subsampling. Improvement threshold $\delta=0.001$ . Vertical bars show one step of the sequential procedure (left to right). Height of the vertical bars represent the number of subsampling iterati

实验结果

研究问题

  • RQ1如何形式化定义并区分基于置换与重拟合的分组特征重要性?
  • RQ2当特征组高度相关时,置换方法与重拟合方法在性能上存在哪些差异?
  • RQ3在高维设置下,顺序程序能否可靠识别稳定且高性能的特征组组合?
  • RQ4组合特征效应图(CFEP)与无监督SPCA相比,在可视化特征组对模型预测的效应方面表现如何?
  • RQ5分组重要性方法在哪些场景下会失效,以及如何缓解这些局限性?

主要发现

  • 在组间相关性较高的场景下,重拟合方法优于置换方法,产生的重要性评分更具意义且更稳定。
  • 顺序分组特征重要性程序成功识别出稳定且高性能的特征组组合,尤其在来自不同数据源的特征组获取成本较高时更具实用性。
  • 所有分组重要性方法(包括基于Shapley与基于置换的方法)在组内特征相关性高度可变时会产生误导性结果,尤其在随机森林模型中更为明显。
  • 特征组中特征数量较多的组通常获得更高的分组重要性评分,但按特征归一化可能导致次优的组选择;因此,GSI应谨慎解读。
  • 当不存在高阶交互作用时,单特征Shapley重要性评分之和等于分组Shapley重要性评分,验证了所提出GSI方法的一致性。
  • 使用监督SPCA的CFEP相比其无监督版本,能生成更具可解释性与意义的可视化结果,尤其在准确捕捉特征组对预测的真实效应方面表现更优。
Figure 2 : Grouped relative importance scores in case of equally sized within-group correlations for random forest (left) and SVM (right). Relative importance is calculated by dividing each of the absolute group importance scores by the importance score of $G_{2}$ . Hence, relative importance of $G_
Figure 2 : Grouped relative importance scores in case of equally sized within-group correlations for random forest (left) and SVM (right). Relative importance is calculated by dividing each of the absolute group importance scores by the importance score of $G_{2}$ . Hence, relative importance of $G_

更好的研究,从现在开始

从阅读论文到最终审阅,大幅缩短您的研究时间。

无需绑定信用卡

本解读由 AI 生成,并经人工编辑审核。