预测脱碳企业的信用评级-论文代码复现

文献来源

【Forecasting credit ratings of decarbonized firms: Comparative assessmentof machine learning models】

文章有代码复现有两个基本工作，1.是提取每个算法的重要性；2.计算每个算法的评价指标

算法有 CRT 分类决策树 ANN 人工神经网络 RFE 随机森林 SVM支持向量机

评价指标有F1 Score ；Specificity ；Accuracy

1.准备数据

分类标签【信誉等级CR1-CR2】和特征向量【变量】

[特征-变量]Probability of Default违约概率
Coverages覆盖范围
Capital Structure资本结构
Liquidity流动性
Profitability盈利能力
Operating Efficiency运营效率
Scale, Scope, and Diversity规模、范围和多样性
Competitive Advantage竞争优势
Fiscal Strength and Credit Conditions财政实力和信用状况
Systemic Governance and Effectiveness系统治理和有效性
Economic Strength经济实力

from sklearn.model_selection import train_test_split
from sklearn.metrics import f1_score, accuracy_score, confusion_matrix
from sklearn.tree import DecisionTreeClassifier
from sklearn.ensemble import RandomForestClassifier
from sklearn.svm import SVC
from sklearn.neural_network import MLPClassifier

# Regenerating the dataset with binary classification
num_companies = 500
features = {
    'Probability_of_Default': np.random.uniform(0, 1, num_companies),
    'Coverages': np.random.uniform(0, 1, num_companies),
    'Capital_Structure': np.random.uniform(0, 1, num_companies),
    'Liquidity': np.random.uniform(0, 1, num_companies),
    'Profitability': np.random.uniform(0, 1, num_companies),
    'Operating_Efficiency': np.random.uniform(0, 1, num_companies),
    'Scale_Scope_and_Diversity': np.random.uniform(0, 1, num_companies),
    'Competitive_Advantage': np.random.uniform(0, 1, num_companies),
    'Fiscal_Strength_and_Credit_Conditions': np.random.uniform(0, 1, num_companies),
    'Systemic_Governance_and_Effectiveness': np.random.uniform(0, 1, num_companies),
    'Economic_Strength': np.random.uniform(0, 1, num_companies),
}
features['Rating'] = np.random.choice([1, 2], num_companies)

# Create a DataFrame
df = pd.DataFrame(features)

# Preview the data
df.head()
df.to_excel(r'C:\Users\12810\Desktop\组会\组会记录\2024-0103寒假集训材料\看论文任务\要讲的\模拟数据.xlsx')

df.head()

	Probability_of_Default	Coverages	Capital_Structure	Liquidity	Profitability	Operating_Efficiency	Scale_Scope_and_Diversity	Competitive_Advantage	Fiscal_Strength_and_Credit_Conditions	Systemic_Governance_and_Effectiveness	Economic_Strength	Rating
0	0.298531	0.328423	0.665525	0.985169	0.107456	0.933863	0.714293	0.681608	0.508398	0.472675	0.508559	2
1	0.841430	0.736388	0.999020	0.689946	0.348265	0.929935	0.066077	0.609516	0.797929	0.048373	0.424858	2
2	0.462897	0.105783	0.716292	0.912855	0.564482	0.850507	0.774066	0.880007	0.737817	0.729397	0.283405	2
3	0.062724	0.073537	0.611761	0.213703	0.483220	0.668749	0.052895	0.924532	0.134043	0.126261	0.910167	1
4	0.242200	0.089723	0.874793	0.659927	0.159241	0.348462	0.828590	0.273572	0.117796	0.154820	0.324018	2

from sklearn.tree import DecisionTreeClassifier
from sklearn.neural_network import MLPClassifier
from sklearn.ensemble import RandomForestClassifier
from sklearn.svm import SVC
from sklearn.metrics import classification_report, accuracy_score
from sklearn.model_selection import train_test_split

# Define features and labels
X = df.drop('Rating', axis=1)
y = df['Rating']

# Split the data into training and test sets
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

# Initialize the classifiers
classifiers = {
    'CRT': DecisionTreeClassifier(random_state=42),
    'ANN': MLPClassifier(random_state=42, max_iter=1000),
    'RFE': RandomForestClassifier(random_state=42),
    'SVM': SVC(random_state=42)
}

# Dictionary to store models' performance metrics
performance_metrics = {}

# Training and evaluating classifiers
for name, clf in classifiers.items():
    clf.fit(X_train, y_train)  # Train the model
    y_pred = clf.predict(X_test)  # Predict on test set
    # Calculate performance metrics
    performance_metrics[name] = classification_report(y_test, y_pred, output_dict=True)
    performance_metrics[name]['accuracy'] = accuracy_score(y_test, y_pred)

# Since ANN and SVM don't provide a direct method for feature importance, we will skip that part as per user's instructions.

performance_metrics

# ... (previous code to generate data and split sets)

# Initialize classifiers
decision_tree = DecisionTreeClassifier(random_state=42)
random_forest = RandomForestClassifier(random_state=42)
ann = MLPClassifier(random_state=42, max_iter=1000)
svm = SVC(random_state=42)

# Train and predict using each classifier
for clf in [decision_tree, random_forest, ann, svm]:
    clf.fit(X_train, y_train)
    y_pred = clf.predict(X_test)
    print(f"{clf.__class__.__name__} metrics:")
    print(classification_report(y_test, y_pred))
    print(f"Accuracy: {accuracy_score(y_test, y_pred)}\n")

# Feature importance for Decision Tree and Random Forest
print(f"Decision Tree Feature Importance: {decision_tree.feature_importances_}")
print(f"Random Forest Feature Importance: {random_forest.feature_importances_}")

D:\install_file\Anaconda3\lib\site-packages\sklearn\neural_network\_multilayer_perceptron.py:691: ConvergenceWarning: Stochastic Optimizer: Maximum iterations (1000) reached and the optimization hasn't converged yet.
  warnings.warn(


DecisionTreeClassifier metrics:
              precision    recall  f1-score   support

           1       0.36      0.38      0.37        45
           2       0.47      0.45      0.46        55

    accuracy                           0.42       100
   macro avg       0.42      0.42      0.42       100
weighted avg       0.42      0.42      0.42       100

Accuracy: 0.42

RandomForestClassifier metrics:
              precision    recall  f1-score   support

           1       0.39      0.40      0.40        45
           2       0.50      0.49      0.50        55

    accuracy                           0.45       100
   macro avg       0.45      0.45      0.45       100
weighted avg       0.45      0.45      0.45       100

Accuracy: 0.45

MLPClassifier metrics:
              precision    recall  f1-score   support

           1       0.40      0.44      0.42        45
           2       0.50      0.45      0.48        55

    accuracy                           0.45       100
   macro avg       0.45      0.45      0.45       100
weighted avg       0.46      0.45      0.45       100

Accuracy: 0.45

SVC metrics:
              precision    recall  f1-score   support

           1       0.45      0.42      0.44        45
           2       0.55      0.58      0.57        55

    accuracy                           0.51       100
   macro avg       0.50      0.50      0.50       100
weighted avg       0.51      0.51      0.51       100

Accuracy: 0.51

Decision Tree Feature Importance: [0.1373407  0.1034318  0.09762338 0.047583   0.13000749 0.04778061
 0.04700612 0.09306039 0.13848067 0.05285945 0.1048264 ]
Random Forest Feature Importance: [0.09758143 0.09809294 0.09033973 0.08525976 0.09331244 0.08281067
 0.094648   0.10032612 0.08863182 0.08533643 0.08366067]


D:\install_file\Anaconda3\lib\site-packages\sklearn\neural_network\_multilayer_perceptron.py:691: ConvergenceWarning: Stochastic Optimizer: Maximum iterations (1000) reached and the optimization hasn't converged yet.
  warnings.warn(

from sklearn.inspection import permutation_importance# For ANN
ann.fit(X_train, y_train)
perm_importance_ann = permutation_importance(ann, X_test, y_test, n_repeats=30, random_state=42)

# For SVM
svm.fit(X_train, y_train)
perm_importance_svm = permutation_importance(svm, X_test, y_test, n_repeats=30, random_state=42)

# Store the permutation importances in a dictionary or DataFrame
feature_importances_ann = perm_importance_ann.importances_mean
feature_importances_svm = perm_importance_svm.importances_mean

# You can then display these importances or further analyze them
print("Feature importances from ANN:", feature_importances_ann)
print("Feature importances from SVM:", feature_importances_svm)

D:\install_file\Anaconda3\lib\site-packages\sklearn\neural_network\_multilayer_perceptron.py:691: ConvergenceWarning: Stochastic Optimizer: Maximum iterations (1000) reached and the optimization hasn't converged yet.
  warnings.warn(


Feature importances from ANN: [-0.04366667 -0.04866667 -0.06166667 -0.01633333 -0.06166667 -0.069
 -0.059      -0.07333333 -0.02866667 -0.041      -0.04266667]
Feature importances from SVM: [-0.002      -0.00633333  0.019       0.035       0.00666667  0.00233333
 -0.00233333 -0.01466667 -0.02        0.00833333  0.02      ]