Abstract: The survival time prediction has been applied on numbers of fields such as clinic medicine, economy and engineering. With the development and broad application of machine learning technology and data mining, lots of survival time prediction models and algorithms have been proposed. Although some have achieved good outcomes, the prediction accuracy still has much space for improvement. This paper proposes a survival time prediction algorithm based on twice supervised learning style. It also realizes the modeling of survival time through both the application of nearest-neighbor algorithm utilized for censored samples estimation and the feature of generalization property of support vector machine (SVM). The experimental results show that the proposed algorithm can achieve precise survival time prediction in clinic, and take the advantage of performance on accurate prognosis.

Key words: survival time prediction, twice supervised learning style, K-nearest neighbor algorithm, support vector machine regression

摘要： 生存时间预测在医学、经济和工程等领域有着广泛的应用。随着机器学习技术和数据挖掘技术的发展和广泛应用，研究人员提出了很多基于机器学习技术的生存时间预测算法。这些算法虽然都取得了良好的效果，但预测精度均有提升的空间。因此，提出了一种基于二次学习风范的生存时间预测算法，并结合最近邻算法在截尾样本估计上的应用以及支持向量机在泛化性能上的优势，实现了对临床生存时间的建模。实验结果表明，该算法能够获取精确的生存时间，且具有预测精度上的性能优势。

关键词: 生存时间预测, 二次监督学习风范, K最近邻算法, 支持向量机回归