Developing a Library of Tree-Based Models for Survival Analysis

A key tool for describing and predicting events is survival analysis, which allows us to predict not only the probability and time of events, but also the change in probability over time. This work presents Survivors, an open-source Python library, that helps solve problems of survival analysis, build individual forecasts of survival and hazard functions, investigate data dependences, assess the quality of forecasts, and conduct experimental studies. The library uses new ways of constructing tree-based models of survival analysis with high sensitivity to real datasets. This work presents a new histogram approach for the splitting of censored data. The models can handle categorial and missing values, cases of informative censorship, and multimodal time distribution. The architecture and components of the library are considered, along with features of the software’s implementation and an experimental comparison to existing libraries of survival analysis.