Implementation:Scikit learn Scikit learn ClassificationThreshold

Knowledge Sources	Scikit_learn Scikit-learn Docs
Domains	Machine Learning, Classification, Model Selection
Last Updated	2026-02-08 15:00 GMT

Overview

Concrete meta-estimator for tuning binary classification decision thresholds provided by scikit-learn.

Description

The _classification_threshold module provides BaseThresholdClassifier and related classes that optimize the decision threshold for binary classifiers. Instead of using the default 0.5 threshold, these meta-estimators search for the threshold that optimizes a given scoring metric (e.g., F1 score, balanced accuracy) via cross-validation. This is particularly useful for imbalanced datasets.

Usage

Use these meta-estimators when you need to optimize the decision threshold of a binary classifier for a specific metric, rather than relying on the default threshold. This is common in scenarios with class imbalance or asymmetric misclassification costs.

Code Reference

Source Location

• Repository: scikit-learn
• File: sklearn/model_selection/_classification_threshold.py

Signature

class BaseThresholdClassifier(ClassifierMixin, MetaEstimatorMixin, BaseEstimator):
    _parameter_constraints: dict = {
        "estimator": [...],
        "response_method": [StrOptions({"auto", "predict_proba", "decision_function"})],
    }

    def __init__(self, estimator, *, response_method="auto"):
        ...

    def fit(self, X, y, **params):
        ...

    def predict(self, X):
        ...

class FixedThresholdClassifier(BaseThresholdClassifier):
    def __init__(self, estimator, *, threshold="auto", response_method="auto"):
        ...

class TunedThresholdClassifierCV(BaseThresholdClassifier):
    def __init__(self, estimator, *, scoring=None, response_method="auto",
                 thresholds=100, cv=None, refit=True, n_jobs=None, random_state=None,
                 store_cv_results=False):
        ...

Import

from sklearn.model_selection import FixedThresholdClassifier, TunedThresholdClassifierCV

I/O Contract

Inputs

Name	Type	Required	Description
estimator	estimator instance	Yes	A binary classifier with predict_proba or decision_function
X	array-like of shape (n_samples, n_features)	Yes	Training input samples
y	array-like of shape (n_samples,)	Yes	Binary target values
scoring	str or callable	No	Scoring metric to optimize the threshold for
cv	int or cross-validator	No	Cross-validation strategy for threshold tuning
threshold	float or "auto"	No	Fixed threshold value (for FixedThresholdClassifier)

Outputs

Name	Type	Description
predictions	ndarray of shape (n_samples,)	Class labels using the optimized threshold
best_threshold_	float	The optimized decision threshold
best_score_	float	The score achieved at the optimal threshold

Usage Examples

Basic Usage

from sklearn.model_selection import TunedThresholdClassifierCV
from sklearn.linear_model import LogisticRegression
from sklearn.datasets import make_classification

X, y = make_classification(n_samples=1000, weights=[0.9, 0.1], random_state=42)
clf = TunedThresholdClassifierCV(
    LogisticRegression(),
    scoring="f1",
    cv=5,
)
clf.fit(X, y)
print(f"Best threshold: {clf.best_threshold_:.3f}")
predictions = clf.predict(X)

Related Pages

• Principle:Scikit_learn_Scikit_learn_Metric_Evaluation