SurvSet: An open-source time-to-event dataset repository
This provides a resource for researchers and practitioners in statistics and machine learning to benchmark time-to-event models, though it is incremental as it focuses on data curation rather than new methods.
The authors tackled the lack of standardized time-to-event datasets for benchmarking machine learning algorithms by creating SurvSet, an open-source repository with 76 consistently formatted datasets, primarily from biomedicine, enabling rapid evaluation of methods.
Time-to-event (T2E) analysis is a branch of statistics that models the duration of time it takes for an event to occur. Such events can include outcomes like death, unemployment, or product failure. Most modern machine learning (ML) algorithms, like decision trees and kernel methods, are supported for T2E modelling with data science software (python and R). To complement these developments, SurvSet is the first open-source T2E dataset repository designed for a rapid benchmarking of ML algorithms and statistical methods. The data in SurvSet have been consistently formatted so that a single preprocessing method will work for all datasets. SurvSet currently has 76 datasets which vary in dimensionality, time dependency, and background (the majority of which come from biomedicine). SurvSet is available on PyPI and can be installed with pip install SurvSet. R users can download the data directly from the corresponding git repository.