В API scikit-learn такого метода нет. Однако вы можете написать это в несколько строк:
import numpy as np
X_train, X_test, y_train, y_test = ... # some dataset
model = ... # some scikit learn model
model.fit(X_train, y_train)
y_pred = model.predict(X_test)
# command to get misclassified examples assuming you have numpy arrays
X_test[np.argwhere(y_pred != y_test)]