A naukri.com initiative
Arxiv
2d
133
Image Credit: Arxiv
Survival Analysis as Imprecise Classification with Trainable Kernels
- Survival analysis is crucial for modeling time-to-event data in various fields like healthcare, engineering, and finance, facing challenges due to censored observations.
- Traditional methods such as the Beran estimator struggle with complex data structures and heavy censoring.
- This paper introduces three new survival models - iSurvM, iSurvQ, and iSurvJ - combining imprecise probability theory and attention mechanisms to handle censored data without parametric assumptions.
- These models represent censored observations using interval-valued probability distributions for each instance over time intervals between events moments.
- They utilize kernel-based Nadaraya-Watson regression with trainable attention weights to compute the imprecise probability distribution over time intervals for the dataset.
- Three decision strategies are considered for training, corresponding to the three proposed models.
- Experiments on synthetic and real datasets show that the proposed models, particularly iSurvJ, outperform the Beran estimator in terms of accuracy and computational complexity.
- Codes for implementing the models are available for public use.
Read Full Article
8 Likes
For uninterrupted reading, download the app