[Day 26] XAI在表格型資料的應用：解析智慧工廠中的鋼材缺陷

import numpy as np
import pandas as pd

url = 'https://raw.githubusercontent.com/andy6804tw/crazyai-xai/main/dataset/stell-faults.csv'
df_data = pd.read_csv(url)

x_feature_names = df_data.columns[:-7].values # 取得特徵欄位名稱
X = df_data[x_feature_names].values # 取出訓練資料特徵
y_label_names = df_data.columns[-7:].values # 取得標籤欄位名稱
y_one_hot_array = df_data[y_label_names].values # 取出標籤
# 使用argmax函數找到每個one hot向量中的最大值索引，並將其視為對應的標籤
y_labels = y_one_hot_array.argmax(axis=1)

print(f'The shape of X: {X.shape}')
print(f'The shape of y_labels: {y_labels.shape}')

The shape of X: (1941, 27)
The shape of y_labels: (1941,)

from sklearn.model_selection import train_test_split

# 切分資料集為訓練集和測試集
X_train, X_test, y_train, y_test = train_test_split(X, y_labels,test_size=0.01, random_state=42, stratify=y_labels)

# 觀察切割後資料的維度與筆數
print(f'The shape of X_train: {X_train.shape}\t y_train: {y_train.shape}')
print(f'The shape of X_test: {X_test.shape}\t y_test: {y_test.shape}')

The shape of X_train: (1746, 27)     y_train: (1746,)
The shape of X_test: (195, 27)       y_test: (195,)

import matplotlib.pyplot as plt

# 查看七種類別筆數
label_counts = pd.Series(y_train).value_counts(sort=False)
fig = label_counts.plot(kind='bar')
fig.set_xticklabels(label_counts)
fig.set_xticklabels(y_label_names)
# 在每個bar上方顯示數值
for index, value in enumerate(label_counts):
    plt.text(index, value, str(value), ha='center', va='bottom')
plt.show()

pip install imbalanced-learn

from imblearn.over_sampling import SMOTE

smo = SMOTE(sampling_strategy='auto',random_state=42)
X_smo, y_smo = smo.fit_resample(X, y_labels)

pip install lightgbm

import lightgbm as lgb

# 建立模型
model = lgb.LGBMClassifier()
# 訓練模型
model.fit(X_train,y_train)

from sklearn.metrics import classification_report


pred_test = model.predict(X_test)
print(classification_report(y_test, pred_test))

              precision    recall  f1-score   support

           0       0.63      0.75      0.69        16
           1       1.00      0.95      0.97        19
           2       0.93      0.95      0.94        39
           3       1.00      0.86      0.92         7
           4       1.00      0.83      0.91         6
           5       0.70      0.67      0.68        39
           6       0.74      0.75      0.75        69

    accuracy                           0.80       195
   macro avg       0.86      0.82      0.84       195
weighted avg       0.80      0.80      0.80       195

import shap
shap.initjs()

# 使用 Kernel SHAP 解釋模型
explainer = shap.KernelExplainer(model=model.predict_proba, data=X_train[:100], link='logit')
shap_values = explainer.shap_values(X=X_test[:10], nsamples=100)

shap.summary_plot(shap_values, X_test, plot_type="bar", feature_names = x_feature_names)

# 觀察測試集中第一筆資料預測的重要程度
index=0
pred_class = int(model.predict(X_test[[index]])[0])
pred_proba = model.predict_proba(X_test[[index]])[0][pred_class]
print(f'測試集第 {index+1} 筆模型預測結果: {pred_class} 機率值: {pred_proba}')
print(f'真實答案: {int(y_test[index])}')
shap.force_plot(explainer.expected_value[pred_class], shap_values[pred_class][index], X_test[index], feature_names=x_feature_names, link='logit')

shap.waterfall_plot(shap.Explanation(values=shap_values[pred_class][index], 
                                    base_values=explainer.expected_value[pred_class], data=X_test[index],  
                                    feature_names=x_feature_names),
                                    max_display=27)

from sklearn.inspection import permutation_importance

# 使用 permutation_importance 函數計算特徵重要性
result = permutation_importance(model, X_test, y_test, n_repeats=10, random_state=42, scoring='accuracy')

# 印出各特徵的平均重要性排序
sorted_idx = result.importances_mean.argsort()[::-1]
for i in sorted_idx:
    print(f"{x_feature_names[i]:<10} importance: {result.importances_mean[i]:.3f} +/- {result.importances_std[i]:.3f}")