Survival Regression with Accelerated Failure Time Model in XGBoost

Survival regression is used to estimate the relation between time-to-event and feature variables, important in application domains such as medicine, marketing, risk management, sales management. Nonlinear tree based machine learning algorithms implemented libraries XGBoost, scikit-learn, LightGBM, CatBoost are often more accurate practice than linear models. However, existing state-of-the-art implementations of tree-based models have offered limited support for survival regression. In this work, we implement loss functions accelerated failure time (AFT) increase modeling different kinds label censoring. We demonstrate with real simulated experiments effectiveness AFT XGBoost respect a number baselines, two respects: generalization performance training speed. Furthermore, take advantage NVIDIA GPUs achieve substantial speedup over multi-core CPUs. To our knowledge, work first implementation that uses processing power GPUs. Starting from 1.2.0 release, package natively supports model. The addition has had significant impact open source community, few statistics packages now use Supplementary materials article available online.