Abstract

The objective of this paper is to provide a new algorithm for the regression of survival data. We propose an algorithm of Survival twin support vector regression (STWSVR), an extension of Twin support vector regression (TWSVR) in binary classification, to explore the analysis of survival data with right censoring. The main algorithm STWSVR is to solve a pair of quadratic programing problems (QPPs) with some tuning parameters. The performance of the algorithm is compared with the survival SVR (SSVR), Cox proportional hazards regression model, Cox proportional hazards model with lasso regularization(Cox-lasso), random survival forests (RSF) and generic gradient boosting algorithm (L2-boosting) on one simulated dataset and 6 practical clinical datasets, using two evaluation measures, C-index and Logrank ${\chi }^2$-statistic.

摘要

本文的目的是为生存数据的回归提供一种新的算法。我们提出了一种生存孪生支持向量回归 (STWSVR) 算法，它是孪生支持向量回归 (TWSVR) 在二元分类中的扩展，以探索右删失生存数据的分析。STWSVR 的主要算法是用一些调整参数来解决一对二次规划问题 (QPP)。该算法的性能与生存 SVR (SSVR)、Cox 比例风险回归模型、带有套索正则化的 Cox 比例风险模型 (Cox-lasso)、随机生存森林 (RSF) 和通用梯度提升算法 (L2-boosting) 进行了比较在一个模拟数据集和 6 个实际临床数据集上，使用两种评估方法，C-index 和 Logrank${\chi }^2$-统计。