Prediction models for disease onset are critical in biomedical research and survival analysis. With machine learning methods increasingly be
Prediction models for disease onset are critical in biomedical research and survival analysis. With machine learning methods increasingly being used to handle survival data with censoring, unbiased transformation theory offers an alternative method for estimating survival tasks in the presence of such censoring, thereby enhancing model accuracy. This study aims to develop a reliable and efficient prediction algorithm that utilizes unbiased transformation to improve machine learning model performance on interval-censored data. Therefore, we present ICBoost, a novel survival algorithm that integrates regression trees and ensemble methods specifically designed for interval-censored data. Unlike right-censored data, where the exact event time is unknown but occurs after a known time point, interval-censored data only provides intervals within which the event occurred. The inherent complexity of interval-censored data poses challenges for accurate survival prediction. To overcome these challenges, we propose a kernel density estimation-based unbiased transformation approach to estimate failure time. Furthermore, we develop a novel ensemble framework that combines XGBoost with censoring unbiased transformation. This framework allows for investigating the relationships between patient survival and covariates, thereby enhancing prediction accuracy and model interpretability. We evaluated the performance of the ICBoost algorithm against existing methods using various real and simulated datasets. Our results showed that ICBoost exhibited superior performance in identifying hidden patterns for survival prediction tasks, particularly in datasets related to Alzheimer’s disease and the emergence of permanent teeth in the medical field. ICBoost outperformed other methods, as evidenced by lower root-mean-square error, mean absolute error, and Brier score values.
