Censoring Unbiased Regression Trees and Ensembles

Area of Science:

• Statistics
• Machine Learning
• Biostatistics

Background:

• Survival analysis is crucial for time-to-event data.
• Existing methods like random survival forests have limitations.
• Novel approaches are needed for robust survival tree and ensemble construction.

Purpose of the Study:

• To propose a novel paradigm for regression trees and ensemble learning in survival analysis.
• To introduce generalizations of CART and Random Forests for survival data.
• To develop new algorithms: Censoring Unbiased Regression Trees and Censoring Unbiased Regression Ensembles.

Main Methods:

• Generalizing CART and Random Forests algorithms for arbitrary loss functions.
• Extending censoring unbiased transformations theory for loss functions.
• Developing and implementing Censoring Unbiased Regression Trees and Ensembles.

Main Results:

• The proposed algorithms, Censoring Unbiased Regression Trees and Ensembles, are developed.
• These methods can be implemented using existing software for squared error loss.
• Simulations and real-data applications show competitive or improved performance compared to existing methods.

Conclusions:

• The new algorithms provide effective tools for survival analysis.
• Censoring Unbiased Regression Ensembles offer a robust alternative for survival prediction.
• These advancements contribute to the field of machine learning in biostatistics.