Misleading plot example for Isolation Forest method

Describe the issue linked to the documentation

I have observed that the plotted feature receive the color of the expected label, i.e., train, test, and outlier. https://github.com/scikit-learn/scikit-learn/blob/778b11904e8ec0286e977582d37e7ca495947ee5/examples/ensemble/plot_isolation_forest.py#L57-L62

Instead, I would expect to plot the predicted labels, i.e., the color of the plotted features according to the prediction of the method. In the published example, the variables defined below are not used for the plot. https://github.com/scikit-learn/scikit-learn/blob/778b11904e8ec0286e977582d37e7ca495947ee5/examples/ensemble/plot_isolation_forest.py#L45-L47

Suggest a potential alternative/fix

The expected plot would be something like this:

b1 = plt.scatter(X_train[:, 0], X_train[:, 1], c=['white' if x > 0 else 'red' for x in y_pred_train],
                 s=20, edgecolor='k')
b2 = plt.scatter(X_test[:, 0], X_test[:, 1], c=['green' if x > 0 else 'red' for x in y_pred_test],
                 s=20, edgecolor='k')
c = plt.scatter(X_outliers[:, 0], X_outliers[:, 1], c=['yellow' if x > 0 else 'red' for x in y_pred_outliers],
                s=20, edgecolor='k')

And the result in this case would vary:

Link to the online example: https://scikit-learn.org/stable/auto_examples/ensemble/plot_isolation_forest.html

[pedrormjunior]

I agree with you guys, to plot a discrete decision boundary would be nice!

[cmarmo]

Related to #22406.