Title: Random splitting random forest for survival analysis with non-functional \& functional covariates in the EEG-fNIRS trial
Authors: Mohammad Fayaz - Shahid Beheshti University of Medical Sciences (Iran) [presenting]
Nezhat Shakeri - Shahid Beheshti University of Medical Sciences (Iran)
Alireza Abadi - Shahid Beheshti University of Medical Sciences (Iran)
Soheila Khodakarim - Shahid Beheshti University of Medical Sciences (Iran)
Abstract: In biostatistics, the survival analysis methods are very popular for analyzing the time-to-event data with censors. The regression tree and random forest are among models that can also handle survival data and we develop a random forest and bagging that can consider the multiple functional covariates with a random-splitting approach. The functional covariate in each tree is randomly split by generating the random number from the exponential distribution and the summary statistics such as average, median, etc. are calculated for these intervals and we put them in the regular algorithm. A new variable importance plot was produced that shows the most important parts of each functional covariate. There are some other functional models such as a functional linear cox regression with Bayesian estimation, optimal estimation, penalized partial likelihood function, joint Bahadur representation of estimators, functional joint models for longitudinal and time-to-event data, functional ensemble survival tree, synthetic data, and additive functional cox models. We applied and compared some of them on a public dataset from the Electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS) trial for the discrimination/selection response (DSR) task. The functional covariates are the event-related potential of EEG and the time-to-event response is the latency of fNIRS for each group.