Kernel machines for current status data

Yael Travis-Lumer, Yair Goldberg

Research output: Contribution to journalArticlepeer-review

Abstract

In survival analysis, estimating the failure time distribution is an important and difficult task, since usually the data is subject to censoring. Specifically, in this paper we consider current status data, a type of data where the failure time cannot be directly observed. The format of the data is such that the failure time is restricted to knowledge of whether or not the failure time exceeds a random monitoring time. We propose a flexible kernel machine approach for estimation of the failure time expectation as a function of the covariates, with current status data. In order to obtain the kernel machine decision function, we minimize a regularized version of the empirical risk with respect to a new loss function. Using finite sample bounds and novel oracle inequalities, we prove that the obtained estimator converges to the true conditional expectation for a large family of probability measures. Finally, we present a simulation study and an analysis of real-world data that compares the performance of the proposed approach to existing methods. We show empirically that our approach is comparable to current state of the art, and in some cases is even better.

Original languageEnglish
Pages (from-to)349-391
Number of pages43
JournalMachine Learning
Volume110
Issue number2
DOIs
StatePublished - Feb 2021
Externally publishedYes

Bibliographical note

Publisher Copyright:
© 2020, The Author(s), under exclusive licence to Springer Science+Business Media LLC, part of Springer Nature.

Keywords

  • Kernel machines
  • Oracle inequalities
  • Support vector regression
  • Survival analysis
  • Universal consistency

ASJC Scopus subject areas

  • Software
  • Artificial Intelligence

Fingerprint

Dive into the research topics of 'Kernel machines for current status data'. Together they form a unique fingerprint.

Cite this