Boruta

What it is and what it does

Boruta is a Python implementation of an all-relevant feature selection algorithm that identifies every feature carrying information useful for prediction, rather than finding a minimal optimal subset. It wraps ensemble methods from scikit-learn and uses a shadow-feature comparison approach with statistical testing to rank and select features. Unlike minimal-optimal methods that depend on classifier choice, Boruta aims to discover all contributing factors to your target variable, making it useful for exploratory analysis and understanding data relationships.

The package provides a scikit-learn-compatible interface (fit, transform, fit_transform) and includes refinements over the original R implementation: automatic estimator selection, feature ranking, percentile-based thresholds instead of strict maximums, and a two-step multiple-testing correction. It depends on numpy, scipy, and scikit-learn, making installation straightforward.

Use it for:

• Exploratory data analysis: identify all features related to your target to understand which factors influence your phenomenon.
• Biological or medical data: use relaxed thresholds (perc parameter) when Bonferroni correction is too harsh for your domain.
• Feature engineering validation: rank candidate features to guide which ones to engineer further or combine.
• Preprocessing before minimal-optimal methods: run Boruta first to reduce noise, then apply stricter selectors.
• Interpretability: obtain feature rankings and support masks to explain model decisions to stakeholders.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Boruta performs all-relevant feature selection by identifying all features carrying predictive information, using ensemble methods with a scikit-learn-compatible interface.

Yes, with conditions. Boruta is well-established (1625 GitHub stars, 258914 monthly downloads) and has no known vulnerabilities. Install it if you need all-relevant feature selection with a scikit-learn-compatible interface and can accept aging maintenance (last release 2024-08-13). Not suitable if you require active, frequent updates or the latest algorithmic refinements.

Install

boruta on PyPI

pip install boruta

uv add boruta

poetry add boruta

Installing Boruta

Before you install

Low friction: pure Python wheel with only three standard dependencies (numpy, scipy, scikit-learn). Last release 2024-08-13; repository is active but maintenance status is aging with 731 days since release.

License in practice

BSD 3 clause is permissive; you can use, modify, and distribute Boruta with minimal restrictions in commercial or private projects.

Quickstart

pip install Boruta

from boruta import BorutaPy

feat_selector = BorutaPy(estimator, n_estimators='auto', verbose=2, random_state=1)
feat_selector.fit(X, y)
X_filtered = feat_selector.transform(X)

Requires numpy arrays as input (not pandas DataFrames directly); convert with .values if needed.

Verify before relying

• Whether the package is actively maintained going forward (aging status suggests infrequent updates).
• Performance characteristics on datasets with thousands of features.
• Current state of the original R package and how closely this implementation tracks it.

Package facts

Evidence: Boruta-0.4.3-py3-none-any.whl

Keywords: feature selection, machine learning, random forest

Tags

More Artificial Intelligence packages