LightGBM

LightGBM is an open-source, high-performance Gradient Boosting Decision Tree (GBDT) framework for efficient, scalable machine learning.

LightGBM, developed by Microsoft, delivers a distributed GBDT framework optimized for speed and low memory consumption on large-scale data. It achieves superior performance—often up to 20 times faster than alternatives like XGBoost—by employing key innovations: a **leaf-wise** tree growth strategy (maximizing loss reduction) instead of the traditional level-wise approach, and a **histogram-based** algorithm for faster split-point finding. Further efficiency comes from **Gradient-based One-Side Sampling (GOSS)** and **Exclusive Feature Bundling (EFB)**, which reduce the number of data instances and features, respectively. The framework supports parallel, distributed, and GPU learning, making it a powerful tool for classification, regression, and ranking tasks.

https://lightgbm.readthedocs.io