Learn AI Create the Future — with Pranay
Main site

Error database

ValueError: The least populated class in y has only 1 member (stratify)

A class in your labels has fewer examples than a stratified split needs. Count the classes first — a 1-member class is usually a label typo, not a real category.

The message you saw
ValueError: The least populated class in y has only 1 member (stratify)

By Updated

The error

Output
ValueError: The least populated class in y has only 1 member, which is too few. The minimum number of groups for any class cannot be less than 2.

Cross-validation raises the sibling:

Output
UserWarning: The least populated class in y has only 3 members, which is less than n_splits=5.

What it means

You asked for a stratified split — one that keeps class proportions equal in train and test. To put a class on both sides, the class needs at least 2 examples. One class has only 1. For k-fold cross-validation, every class needs at least n_splits examples, one per fold.

Why it happens

Look at the class counts before blaming the splitter:

python
print(y.value_counts())
Output
churned        4890
active         5093
churned           1

A class with 1 member is very often a typo or formatting slip — "churned " with a trailing space, "Active" versus "active", an ID that leaked into the label column. Genuine rare classes also happen: a disease with 3 cases, a product bought once.

How to fix it

1. Fix label noise first. If the rare class is a typo, normalise:

python
y = y.str.strip().str.lower()

This usually deletes the phantom class and the error with it.

2. Merge genuinely rare classes into a meaningful bucket.

python
counts = y.value_counts()
rare = counts[counts < 10].index
y = y.replace(dict.fromkeys(rare, "other"))

Do this only when "other" makes sense for your question.

3. Drop classes too rare to learn from — and say so in your write-up.

python
keep = y.isin(counts[counts >= 10].index)
X, y = X[keep], y[keep]

A model cannot learn a class from 2 examples anyway.

4. For cross-validation, reduce the fold count.

python
cross_val_score(model, X, y, cv=3)     # smallest class has 3 members

5. Drop stratify only as a last resort. Without it, a rare class can land entirely in the test set — or entirely outside it — and your metrics become a lottery. Prefer fixes 1-4.

How to prevent it

Print y.value_counts() as the first act of any classification project. It surfaces typos, class imbalance and impossible classes in one line, before they become errors deeper in the pipeline. Re-check after every join or filter that touches the label column.