fmrs

fmrs obtains maximum likelihood and maximum penalized likelihood estimation for Finite Mixture of Regression (FMR) and Finite Mixture of Accelerated Failure Time Regression (FMAFTR) models via the Expectation-Maximization algorithm and Newton-Raphson algorithm, with component-wise tuning parameter selection for variable selection.

Key Features:

• Maximum Likelihood Estimation for FMR Models: Obtains MLE for Finite Mixture of Regression models with Normal distribution via the Expectation-Maximization algorithm.
• Maximum Likelihood Estimation for FMAFTR Models: Obtains MLE for Finite Mixture of Accelerated Failure Time Regression models subject to right censoring with Log-Normal distribution via the EM algorithm and with Weibull distribution via the EM algorithm and Newton-Raphson algorithm.
• Maximum Penalized Likelihood Estimation: Obtains MPLE for both FMR and FMAFTR models collectively called FMRs.
• Component-wise Tuning Parameter Selection: Implements component-wise BIC-based tuning parameter selection.
• Regularization Methods: Provides Ridge Regression and Elastic Net.

Scientific Applications:

• Variable Selection in Mixture Regression: Enables variable selection in Finite Mixture of AFT Regression and FMR models through penalized likelihood estimation.
• Survival Analysis with Right Censoring: Supports analysis of censored time-to-event data using Finite Mixture of Accelerated Failure Time Regression models with Log-Normal and Weibull distributions.

Methodology:

Parameter estimation uses the Expectation-Maximization algorithm for FMR models with Normal distribution and for FMAFTR models with Log-Normal distribution; the Newton-Raphson algorithm is used alongside EM for FMAFTR models with Weibull distribution. Maximum penalized likelihood estimation employs Ridge Regression and Elastic Net penalties with component-wise BIC for tuning parameter selection.