Study
PLCO
(Learn more about this study)
Project ID
PLCO-453
Initial CDAS Request Approval
Feb 11, 2019
Title
Feature Selection for Survival Analysis with Competing Risks using Deep Learning
Summary
Deep learning models for survival analysis have gained significant attention in the literature, but they suffer from severe performance deficits when the dataset contains many irrelevant features. We will give empirical evidence for this problem in real-world medical settings using the state-of-the-art model DeepHit. Furthermore, we will develop methods to improve the deep learning model through novel approaches to feature selection in survival analysis. We will propose filter methods for hard feature selection and a neural network architecture that weights features for soft feature selection. Our experiments on a real-world medical dataset will demonstrate that substantial performance improvements against the original models are achievable.
Aims
Recent research has produced a variety of successful new deep learning models for survival analysis. Whilst some methods have strong parametric assumptions, more general models have been developed. However, deep learning approaches suffer from performance deficits when there are many irrelevant features. This can certainly be the case in medical datasets, where numerous features may be recorded about a patient. In this paper, we give evidence for this problem using DeepHit on a large real-world medical dataset, and propose feature selection techniques to achieve substantial performance improvements.
Collaborators
Carl Rietschel, University of Oxford, United Kingdom
Jinsung Yoon, University of California, Los Angeles, USA
Mihaela van der Schaar, University of California, Los Angeles, USA
