Introduction

ImportError: No module named sklearn.cross_validation usually means your code was written for an older scikit-learn API. Modern scikit-learn moved those utilities into sklearn.model_selection, so the fix is usually small once you know where the feature lives now.

Why the Import Fails

Older tutorials often import helpers like train_test_split or KFold from sklearn.cross_validation. That module was deprecated and then removed as scikit-learn reorganized model evaluation tools. The functionality still exists, but the package path changed.

This kind of error is common when you copy code from blog posts, notebooks, or older internal examples. The rest of the program may still be valid; the import line is the part that no longer matches the installed library version.

The Correct Replacement

Most code should switch from sklearn.cross_validation to sklearn.model_selection.

1from sklearn.datasets import load_iris
2from sklearn.linear_model import LogisticRegression
3from sklearn.model_selection import train_test_split, cross_val_score
4
5X, y = load_iris(return_X_y=True)
6
7X_train, X_test, y_train, y_test = train_test_split(
8    X,
9    y,
10    test_size=0.2,
11    random_state=42,
12    stratify=y,
13)
14
15model = LogisticRegression(max_iter=200)
16model.fit(X_train, y_train)
17
18scores = cross_val_score(model, X, y, cv=5)
19print(scores)
20print("Mean accuracy:", scores.mean())

If your original code used KFold, StratifiedKFold, ShuffleSplit, or cross_val_score, the new import path is usually the same pattern: import the symbol from sklearn.model_selection.

How to Confirm What Version You Have

Before changing more code than necessary, confirm the installed scikit-learn version. That tells you whether the issue is a stale import, an outdated environment, or both.

python -c "import sklearn; print(sklearn.__version__)"

If the version is modern, update the import. If the environment is intentionally pinned to a legacy dependency set, the older import may still work there, but keeping new code on removed modules is usually a bad tradeoff. It makes upgrades harder and creates confusion for anyone who expects current package layouts.

Updating a Larger Codebase Safely

In a real project, do not stop at fixing one file. Search for all references to the deprecated module and replace them consistently.

from sklearn.model_selection import KFold, GridSearchCV, train_test_split

Then rerun tests or training notebooks. Import errors are often only the first visible symptom of an outdated example. You may also find renamed parameters, changed defaults, or stricter validation in newer scikit-learn versions.

When Downgrading Is the Wrong Fix

Developers sometimes solve the error by installing a very old version of scikit-learn. That can make the import work, but it usually creates a worse maintenance problem. Older releases may depend on legacy NumPy versions, fail on newer Python interpreters, or behave differently from the rest of your team’s environment.

Downgrading is only reasonable when you are reproducing an old experiment that truly requires the original stack. For active projects, updating the import path is the cleaner fix.

Common Pitfalls

• Patching only one import line is not enough if the project uses the old module in many files. Search the codebase before closing the issue.
• Mixing tutorials from different scikit-learn eras can produce inconsistent examples. Check the publication date of outside material.
• Downgrading scikit-learn to avoid updating imports often introduces dependency conflicts with Python, NumPy, or pandas.
• Assuming every old symbol moved to the exact same location can waste time. Most moved to model_selection, but verify unusual utilities individually.
• Forgetting to rerun tests after the import fix can hide other compatibility issues.

Summary

• 'sklearn.cross_validation is an old module path that no longer exists in modern scikit-learn.'
• Replace those imports with equivalents from sklearn.model_selection.
• Check the installed scikit-learn version before deciding whether to upgrade code or reproduce a legacy environment.
• Avoid downgrading the library unless you are intentionally preserving an old stack.
• After updating imports, run the project again to catch any additional API changes.