【XAI】评估XAI方法与实践应用

一、XAI评估指标示例代码开发

1. 忠诚度（Fidelity）示例

忠诚度衡量解释与模型真实决策的一致性，常用输入扰动法（Perturbation-based）实现。示例中通过逐步遮蔽重要特征，观察模型输出变化。

python

解释

import numpy as np from sklearn.ensemble import RandomForestClassifier from sklearn.datasets import load_iris from sklearn.model_selection import train_test_split # 加载数据与训练模型 data = load_iris() X_train, X_test, y_train, y_test = train_test_split(data.data, data.target, random_state=42) model = RandomForestClassifier(random_state=42) model.fit(X_train, y_train) # 简单特征重要性排序（用模型自带的feature_importances_） feature_importances = model.feature_importances_ sorted_idx = np.argsort(feature_importances)[::-1] def fidelity_perturbation(mod