Mixture of Censored Partially Linear Experts Models for Symmetric or Asymmetric Data with or without Outlying Observations: A novel approach with two-piece scale-mixture of normal distributions
Abstract
Finite mixture of partially linear model and finite mixture of experts models offer flexible tools for analyzing heterogeneous data, especially when covariates influence subgroup structure. Partially linear models further enhance this flexibility by combining linear and nonlinear effects. In many applications, however, censoring and departures from normality complicate estimation. To address these issues, we introduce a semiparametric mixture of censored partially linear expert models based on the two-piece scale mixture of normal family, which can accommodate skewness, heavy tails, and censoring. The proposed framework enables closed-form expressions in the E-step of an ECME algorithm, avoiding MCMC-based approximations and yielding stable maximum likelihood estimates. Simulation studies and real data analyses demonstrate the effectiveness of the proposed method in settings with both heterogeneity and censoring.
Related
- Multi-Absorbing Phase-Type Distributions for Right-Censored Competing Risks Data
- M-Eigenpairs of Partially Symmetric Tensors: Exact Reformulation and Perturbation Bounds
- A Stochastic Maximum Principle for Partially Observed Jump-Diffusion Systems with State-Dependent Counting-Process Observations
- Approximate maximum likelihood estimation for partially observed fractional diffusions from discrete observations
- Joint distributions of eigenvectors of symmetric random tensors
BibTeX
@misc{maleki2026mixture,
title = {{Mixture of Censored Partially Linear Experts Models for Symmetric or Asymmetric Data with or without Outlying Observations: A novel approach with two-piece scale-mixture of normal distributions}},
author = {Mohsen Maleki and Geoffrey J. McLachlan},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.27424},
eprint = {2610.27424},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.27424},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Mixture of Censored Partially Linear Experts Models for Symmetric or Asymmetric Data with or without Outlying Observations: A novel approach with two-piece scale-mixture of normal distributions
AU - Mohsen Maleki
AU - Geoffrey J. McLachlan
PY - 2026
DA - 2026/10/20
PB - arXiv
DO - 10.48550/arxiv.2610.27424
UR - https://arxiv.org/abs/2610.27424
AB - Finite mixture of partially linear model and finite mixture of experts models offer flexible tools for analyzing heterogeneous data, especially when covariates influence subgroup structure. Partially linear models further enhance this flexibility by combining linear and nonlinear effects. In many applications, however, censoring and departures from normality complicate estimation. To address these issues, we introduce a semiparametric mixture of censored partially linear expert models based on the two-piece scale mixture of normal family, which can accommodate skewness, heavy tails, and censoring. The proposed framework enables closed-form expressions in the E-step of an ECME algorithm, avoiding MCMC-based approximations and yielding stable maximum likelihood estimates. Simulation studies and real data analyses demonstrate the effectiveness of the proposed method in settings with both heterogeneity and censoring.
ER -