A1255
Title: Incomplete trajectories with covariates
Authors: Thi Bao Tram Ngo - University Evry Paris-Saclay (France) [presenting]
Juhyun Park - ENSIIE (France)
Abstract: Functional data are often observed only over short, disconnected portions of their underlying trajectories, which makes the recovery of complete curves a challenging task, particularly in the presence of heterogeneity across subjects. Two approaches for reconstructing incomplete functional trajectories while incorporating covariate information are presented. The first is based on a Markov chain description of local trajectory evolution, with transition probabilities estimated by smoothing over time, state space, and covariates. The second relies on conditional quantiles of subject-specific slope summaries given the current level, which are then used to propagate trajectories through a covariate-dependent dynamic reconstruction scheme. Both methods are intended for settings in which each subject is observed only over a limited interval and where continuous or discrete covariates may account for part of the between-subject variability. Methodological ideas, the choice of tuning parameters, and the construction of prediction bands are discussed. Simulation studies illustrate the performance of the two approaches in terms of reconstruction accuracy and robustness. The comparison highlights their complementary strengths and offers practical insights into the reconstruction of fragmented functional data in the presence of covariates.
