Description
Semi-Supervised Learning under a Mixed-Missingness Mechanism in Finite Mixture Models.
Description
Implements a semi-supervised learning framework for finite mixture models under a mixed-missingness mechanism. The approach models both missing completely at random (MCAR) and entropy-based missing at random (MAR) processes using a logistic–entropy formulation. Estimation is carried out via an Expectation–-Conditional Maximisation (ECM) algorithm with robust initialisation routines for stable convergence. The methodology relates to the statistical perspective and informative missingness behaviour discussed in Ahfock and McLachlan (2020) <doi:10.1007/s11222-020-09971-5> and Ahfock and McLachlan (2023) <doi:10.1016/j.ecosta.2022.03.007>. The package provides functions for data simulation, model estimation, prediction, and theoretical Bayes error evaluation for analysing partially labelled data under a mixed-missingness mechanism.