CatBoost

Developed at Yandex, CatBoost implements gradient boosting over decision trees and is best known for handling categorical features natively, so users can skip manual one-hot or target encoding. Its ordered boosting scheme counters the target leakage that causes overfitting in conventional gradient boosting, and the library provides fast out-of-the-box GPU and multi-GPU training along with quick CPU inference. Interfaces exist for Python, R, and the command line, with Apache Spark integration for distributed learning, and the toolkit bundles visualization, cross-validation, parameter tuning, and feature importance analysis for model interpretation. CatBoost performs competitively with XGBoost and LightGBM on many tabular datasets and covers classification, regression, and ranking objectives, with model export to several runtimes for deployment. Released under the Apache 2.0 license, it is widely used by data scientists working on structured business data where categorical columns dominate.