Machine Learning for Survival Analysis

Survival analysis is a mature field with decades of methodological development, yet machine learning survival analysis is still taking shape as a discipline in its own right. While machine learning methods for time-to-event prediction are increasingly used in health care, clinical research, actuarial science, engineering, and industry, a critical gap remains in the literature: few texts bridge the theoretical foundations of survival analysis with the methods, workflows, and evaluation tools of modern machine learning. Machine Learning in Survival Analysis fills this gap by providing a systematic treatment of machine learning approaches to time-to-event prediction.

From nonparametric estimators, Cox proportional hazards models and parametric models to random forests, support vector machines, gradient boosting machines, and neural networks, this book offers a rigorous yet accessible journey through the field. It formally defines the survival analysis machine learning task, introduces key prediction targets, and explains how censoring and truncation change the structure of standard predictive modeling problems. Beyond model fitting, the book gives detailed attention to model evaluation, including discrimination, calibration, scoring rules, censoring adjustment, and the assumptions required for valid comparison. The book covers single-event right-censored data, with extensions to other censoring mechanisms, truncation, competing risks, reduction methods, and event history analysis more generally.

Key Features

• Comprehensive coverage from survival analysis foundations to modern machine learning methods, including non-parametric, semi-parametric, and fully parametric models, random forests, support vector machines, gradient boosting machines, and neural networks
• Formal treatment of survival analysis as a machine learning task, with clear definitions of censoring, truncation, competing risks, and multiple prediction targets
• Detailed evaluation framework covering discrimination, calibration, scoring rules, censoring adjustment, and the practical limitations of commonly used measures
• Coverage of reduction methods that connect survival analysis to standard regression and classification frameworks
• Extensions beyond single-event right censoring, including interval censoring, truncation, competing risks, and event history analysis

This book is designed for graduate students, researchers, and data science practitioners with foundational knowledge of statistics and machine learning. It serves as a textbook for courses in survival analysis, biostatistics, machine learning, and applied predictive modeling, as well as a reference for practitioners developing, evaluating, or deploying survival models in clinical, industrial, engineering, financial, or scientific settings. Whether the reader is learning survival analysis for the first time, moving from standard machine learning to time-to-event modeling, or seeking a unified treatment of classical and modern approaches, this book provides the theory needed to understand, evaluate, and apply machine learning methods to time-to-event data and temporal processes with partially observed information.