Information Geometry for Survival Analysis and Feature Selection by Neural Networks

Antonio Eleuteri, Roberto Tagliaferri, Leopoldo Milano, Michele de Laurentiis

Research output: Chapter in Book/Report/Conference proceedingChapter

Abstract

In this chapter, an information geometric approach to survival analysis has been described. It shows how a neural network can be used to model the probability of failure of a system, and how it can be trained by minimizing a suitable divergence functional in a Bayesian framework. By using the trained network, minimization of the same divergence functional allows for fast, efficient, and exact feature selection. The performance of the algorithms has been illustrated on some datasets. In this chapter, a novel approach to survival analysis has been presented. Neural network architecture has been defined and trained according to information geometric principles. The same concepts are applied to exploit the geometric structure of the network and an algorithm is formulated which efficiently solves the feature selection problem. The proposed approach does not make any assumption on the form of the survival process, and does not use discretizations or piecewise approximations as other neural network approaches to survival analysis do.

Original languageEnglish
Title of host publicationOutcome Prediction in Cancer
PublisherElsevier
Pages171-189
Number of pages19
ISBN (Print)9780444528551
DOIs
Publication statusPublished - 2007

ASJC Scopus subject areas

  • Biochemistry, Genetics and Molecular Biology(all)

Fingerprint Dive into the research topics of 'Information Geometry for Survival Analysis and Feature Selection by Neural Networks'. Together they form a unique fingerprint.

Cite this