Random jump forests estimate multi-state transition rates without Markov assumption
Original title: Random Jump Forests
One tree partitions covariates for all transitions, with Nelson–Aalen estimates and an R package.
Random jump forests are introduced to estimate conditional cumulative transition rates and occupation probabilities in multi-state models. A single tree partitions the covariate space for all transitions using partially observed trajectories and Nelson–Aalen estimators at terminal nodes. Occupation probabilities use the product integral and do not require the Markov assumption. Simulations approach correctly specified benchmarks and improve when those benchmarks are misspecified; the method is in the R package JumpForests.
- Uniform consistency and a functional central limit theorem assume finitely many covariate values.
- Pointwise intervals and simultaneous bands are available in the Markov case.
- Coverage bands were near nominal; pooling heterogeneous rates can undercover.
- Illustration uses four biomedical data sets plus variable importance.
From the original article
1 min read at arxiv.orgRandom jump forests are introduced as a nonparametric tool for estimating conditional cumulative transition rates and occupation probabilities in multi-state models. The main idea is that a single tree partitions the covariate space for all transitions simultaneously, using partially observed trajectories, and with Nelson--Aalen estimators at the terminal nodes.
By Martin Bladt · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.