Implementation:Scikit learn contrib Imbalanced learn geometric mean score

Knowledge Sources	imbalanced-learn imbalanced-learn Docs
Domains	Machine_Learning, Model_Evaluation, Imbalanced_Learning
Last Updated	2026-02-09 03:00 GMT

Overview

Concrete tool for computing the geometric mean of per-class sensitivities provided by the imbalanced-learn library.

Description

The geometric_mean_score function computes the G-mean metric. It supports binary, multiclass, macro, weighted, and micro averaging. The correction parameter substitutes a non-zero value for classes with zero sensitivity to prevent the entire metric from collapsing. Uses scipy.stats.gmean internally.

Usage

Import this function to evaluate classifier performance on imbalanced datasets using the geometric mean metric.

Code Reference

Source Location

• Repository: imbalanced-learn
• File: imblearn/metrics/_classification.py
• Lines: L550-736

Signature

def geometric_mean_score(
    y_true,
    y_pred,
    *,
    labels=None,
    pos_label=1,
    average="multiclass",
    sample_weight=None,
    correction=0.0,
):
    """
    Args:
        y_true: array-like of shape (n_samples,) - Ground truth labels.
        y_pred: array-like of shape (n_samples,) - Predicted labels.
        labels: array-like or None - Label subset to include.
        pos_label: int - Positive class for binary (default: 1).
        average: str - Averaging method: 'multiclass', 'binary', 'macro',
            'weighted', 'micro', None (default: 'multiclass').
        sample_weight: array-like or None - Sample weights.
        correction: float - Substitute for zero sensitivity (default: 0.0).
    Returns:
        float or ndarray - Geometric mean score.
    """

Import

from imblearn.metrics import geometric_mean_score

I/O Contract

Inputs

Name	Type	Required	Description
y_true	array-like of shape (n_samples,)	Yes	Ground truth labels
y_pred	array-like of shape (n_samples,)	Yes	Predicted labels
average	str or None	No	Averaging strategy (default: 'multiclass')
correction	float	No	Substitute for zero sensitivity (default: 0.0)

Outputs

Name	Type	Description
score	float or ndarray	G-mean score (0.0 to 1.0); array if average=None

Usage Examples

from imblearn.metrics import geometric_mean_score

y_true = [0, 1, 2, 0, 1, 2]
y_pred = [0, 2, 1, 0, 0, 1]

# Default multiclass
score = geometric_mean_score(y_true, y_pred)
print(f"G-mean: {score:.3f}")

# Per-class scores
per_class = geometric_mean_score(y_true, y_pred, average=None)
print(f"Per-class: {per_class}")

# With correction for zero sensitivity
corrected = geometric_mean_score(y_true, y_pred, correction=0.001)

Related Pages

Implements Principle

• Principle:Scikit_learn_contrib_Imbalanced_learn_Geometric_Mean_Evaluation

Requires Environment

• Environment:Scikit_learn_contrib_Imbalanced_learn_Python_Scikit_learn