Deep Learning for Survival Analysis

Explore deep learning models for survival analysis in this comprehensive 59-minute lecture by Louise Ferbach, a Kaggle Competitions Master and Actuary Data Scientist at SCOR. Delve into time-to-event prediction problems applicable to credit default, machine failure, and cancer relapse scenarios. Learn about the general framework of survival analysis, including censoring and truncation concepts, and understand why censored data should not be discarded. Examine the Cox regression model, its covariates, and nonparametric baseline hazard. Discover cutting-edge deep learning models like DeepServe and CoxTime, and their application to real-world datasets such as the Democracy Data Set. Gain insights into customer analytics, probability estimation, and binary classification in the context of survival analysis. Explore the use of Electronic Health Record (EHR) databases and evaluation techniques for survival models. This lecture provides a thorough overview of deep learning applications in survival analysis, equipping you with valuable knowledge for various time-to-event prediction challenges.

Introduction
Overview
Survival Analysis
General Framework
Censoring and Truncation
Can we throw away censored data
Cox regression model
Covariates
Nonparametric baseline hazard
Deep learning models
DeepServe
CoxTime
Democracy Data Set
Dips Off
Perspectives
Resources
Customer Analytics
Probability
Binary Classification
EHR Database
Cox Time
Evaluation