Loading timeline…
20162010s
XGBoost
The gradient-boosting implementation that won competition after competition — and still wins on tabular data.
Why It Was Important
Chen and Guestrin's library made gradient boosting fast, regularised and easy to run in parallel, and the effect on applied machine learning was immediate: XGBoost appeared in the winning solution of most Kaggle competitions on structured data for years. It matters historically because it marks the boundary of the deep-learning revolution — on tabular business data, boosted trees kept beating neural networks throughout the 2010s, and largely still do.
Who Invented It
Tianqi Chen, Carlos Guestrin
Developed at the University of Washington and released as open source.
Applications
- Tabular Prediction
- Fraud Detection
- Ranking
- Risk Modelling
Key Papers
- XGBoost: A Scalable Tree Boosting System
Tianqi Chen, Carlos Guestrin · KDD 2016