Finite Mixture Modeling via REBMIX

The REBMIX algorithm for fitting finite mixture models implemented in R package rebmix is presented. It provides functions for random univariate and multivariate finite mixture generation, the number of components, component weights and component parameter estimation, bootstrapping and the plotting of finite mixtures. It requires preprocessing of observations, information criterion and conditionally independent normal, lognormal, Weibull, gamma, binomial, Poisson or Dirac component densities. The algorithm optimizes the component parameters, mixing weights and number of components successively based on boundary conditions, such as the maximum number of components and number of bins or nearest neighbours. The algorithm is robust, time efficient and can be used either to assess an initial set of unknown parameters and number of components, e.g., for the EM algorithm, or as a standalone algorithm providing a good compromise between parametric and nonparametric methods of finite mixture estimation. Univariate and multivariate datasets are analysed for validation purposes. Keywords-Continuous Variable; Discrete Variable; Mixture Estimation; R Package