XGBoost

Introduced in a 2016 KDD paper by Tianqi Chen with Carlos Guestrin at the University of Washington, this C++ library builds ensembles of decision trees by gradient boosting and remains a default pick for tabular prediction. It handles regression, classification, learning-to-rank and survival analysis, accepts custom objectives and metrics, treats missing values natively, and supports categorical features, monotonic and interaction constraints, multi-output targets and DART or random forest modes. Bindings cover Python (with a scikit-learn style estimator), R, the JVM via XGBoost4J, C, C++, Julia, Ruby and Swift. Training scales from a laptop to clusters through Dask, Spark, PySpark, Ray and Kubernetes, and an external memory mode handles datasets bigger than RAM. Setting device to cuda moves histogram tree construction onto an NVIDIA GPU. Install from PyPI, conda-forge or CRAN; releases arrive often, with the 3.4 series shipping in 2026. Apache-2.0 licensing allows unrestricted commercial use. Data scientists working with logs, transactions or feature tables usually try it before reaching for deep learning.