Why does the Confusion Matrix in Soft Voting vary across multiple iterations?

Question

Why does the Confusion Matrix in Soft Voting vary across multiple iterations? Does probability=True have an influence? How can the Confusion Matrix results be consistent in Soft Voting?

from sklearn.model_selection import cross_val_score, KFold
from sklearn.metrics import confusion_matrix
from sklearn.svm import SVC
from sklearn.ensemble import VotingClassifier
import numpy as np
from sklearn.metrics import classification_report

X = dataset_baru.iloc[:, :-1].values
y = dataset_baru.iloc[:, -1].values

# Initialize SVM models with different kernels
svm_kernel1_soft = SVC(kernel='linear', probability=True)
svm_kernel2_soft = SVC(kernel='poly', probability=True)
svm_kernel3_soft = SVC(kernel='rbf', probability=True)

svm_kernel1_hard = SVC(kernel='linear')
svm_kernel2_hard = SVC(kernel='poly')
svm_kernel3_hard = SVC(kernel='rbf')

# Perform K-Fold Cross Validation
kfold = KFold(n_splits=5, shuffle=True, random_state=42)

confusion_matrices_hard = []
confusion_matrices_soft = []

for train_index, test_index in kfold.split(X):
    X_train, X_test = X[train_index], X[test_index]
    y_train, y_test = y[train_index], y[test_index]

    # Initialize VotingClassifier with hard voting
    voting_clf_hard = VotingClassifier(
        estimators=[('svm1', svm_kernel1_hard), ('svm2', svm_kernel2_hard), ('svm3', svm_kernel3_hard)],
        voting='hard'
    )

    # Fit the VotingClassifier with hard voting
    voting_clf_hard.fit(X_train, y_train)

    # Predict labels using hard voting
    y_pred_hard = voting_clf_hard.predict(X_test)



    # Initialize VotingClassifier with soft voting
    voting_clf_soft = VotingClassifier(
        estimators=[('svm1', svm_kernel1_soft), ('svm2', svm_kernel2_soft), ('svm3', svm_kernel3_soft)],
        voting='soft'
    )

    # Fit the VotingClassifier with soft voting
    voting_clf_soft.fit(X_train, y_train)

    # Predict labels using soft voting
    y_pred_soft = voting_clf_soft.predict(X_test)

      # Calculate confusion matrices for hard voting
    confusion_matrices_hard.append(confusion_matrix(y_test, y_pred_hard))

    # Calculate confusion matrices for soft voting
    confusion_matrices_soft.append(confusion_matrix(y_test, y_pred_soft))

# Print classification reports for hard voting
print("Classification Report - Hard Voting:")
for i, cm in enumerate(confusion_matrices_hard):
    print(f"Fold {i+1}:")
    print(classification_report(y_test, y_pred_hard))

# Print classification reports for soft voting
print("Classification Report - Soft Voting:")
for i, cm in enumerate(confusion_matrices_soft):
    print(f"Fold {i+1}:")
    print(classification_report(y_test, y_pred_soft))


# Mencetak laporan klasifikasi untuk hard voting
print("Classification Report - Hard Voting:")
print(classification_report(y_test, y_pred_hard))

# Mencetak laporan klasifikasi untuk soft voting
print("Classification Report - Soft Voting:")
print(classification_report(y_test, y_pred_soft))

When I tried to remove probability=True from SVM, an error occurred.

AttributeError: predict_proba is not available when probability=False