|
|
|
 |
Search published articles |
 |
|
Showing 15 results for Em Algorithm
Hossein Baghishani, Mohammad Mahdi Tabatabaei,
Volume 1, Issue 1 (9-2007)
Abstract
In parameter driven models, the main problem is likelihood approximation and also parameter estimation. One approach to this problem is to apply simpler likelihoods such as composite likelihood. In this paper, we first introduce the parameter driven models and composite likelihood and then define a new model selection criterion based on composite likelihood. Finally, we demonstrate composite likelihood's capabilities in inferences and accurate model selection in parameter driven models throughout a simulation study.
Rahman Farnoosh, Afshin Fallah, Arezoo Hajrajabi,
Volume 2, Issue 2 (2-2009)
Abstract
The modified likelihood ratio test, which is based on penalized likelihood function, is usually used for testing homogeneity of the mixture models. The efficiency of this test is seriously affected by the shape of penalty function that is used in penalized likelihood function. The selection of penalty function is usually based on avoiding of complexity and increasing tractability, hence the results may be far from optimality. In this paper, we consider a more general form of penalty function that depends on a shape parameter. Then this shape parameter and the parameters of mixture models are estimated by using Bayesian paradigm. It is shown that the proposed Bayesian approach is more efficient in comparison to modified likelihood test. The proposed Bayesian approach is clearly more efficient, specially in nonidentifiability situation, where frequentist approaches are almost failed.
Mohamad Babazadeh, Sadegh Rezaee, Mousa Abdi,
Volume 6, Issue 1 (8-2012)
Abstract
In this paper, a new three-parameter lifetime distribution is introduced by combining an extended exponential distribution with a logarithmic distribution. This flexible distribution has increasing, decreasing and upside-down bathtub failure rate shapes. Various properties of the proposed distribution are discussed. The estimation of the parameters attained by EM algorithm and their asymptotic variance and covariance are obtained. In order to assess the accuracy of the approximation of variance and covariance of the maximum likelihood estimator, a simulation study is presented to illustrate the properties of distribution.
Afshin Fallah, Mahsa Nadifar, Ramin Kazemi,
Volume 7, Issue 1 (9-2013)
Abstract
In this paper the regression analysis with finite mixture bivariate poisson response variable is investigated from the Bayesian point of view. It is shown that the posterior distribution can not be written in a closed form due to the complexity of the likelihood function of bivariate Poisson distribution. Hence, the full conditional posterior distributions of the parameters are computed and the Gibbs algorithm is used to sampling from posterior distributions. A simulation study is performed in order to assess the proposed Bayesian model and its efficiency in estimation of the parameters is compared with their frequentist counterparts. Also, a real example presented to illustrate and assess the proposed Bayesian model. The results indicate to the more efficiency of the estimators resulted from Bayesian approach than estimators of frequentist approach at least for small sample sizes.
Nasrin Moradi, Abdolreza Sayyareh, Hanieh Panahi,
Volume 8, Issue 1 (9-2014)
Abstract
In this article, the parameters of the Exponentiated Burr type III distribution have been estimated based on type II censored data using maximum likelihood method with EM algorithm and Bayesian approach under Gamma prior distributions against the squared error, linex and entropy loss functions. Importance sampling technique and Lindley's approximation method have been applied to evaluate these Bayes estimates. The results are checked by simulation study and analyzing real data of acute myelogeneous disease. The Bayes estimates are, generally, better than the MLEs and all estimates improve by increasing sample size.
Aref Khanjari Idenak, Mohammadreza Zadkarami, Alireza Daneshkhah,
Volume 8, Issue 2 (3-2015)
Abstract
In this paper a new compounding distribution with increasing, decreasing, bathtub shaped and unimodal hazard rate function is proposed. The new four-parameters distribution is a generalization of the complementary exponential power distribution. The raw-moments, density function of the order statistics, survival function, hazard rate function, quantiles, mean residual lifetime and reliability function are presented. The estimation of the new distribution in a special case Poisson complementary exponential power distribution is studied by the method of maximum likelihood and EM algorithm. Expression for asymptotic distribution for the maximum likelihood estimation of the parameters of the PCEP distribution are obtained and for determining the precision of the variance and covariance of the estimations, a simulation is used, Then experimental results are illustrated based on the real data set.
Maliheh Heidari, Farzad Eskandari,
Volume 11, Issue 1 (9-2017)
Abstract
In this paper the issue of variable selection with new approach in finite mixture of semi-parametric regression models is studying, although it is supposed that data have Poisson distribution. When we use Poisson distribution, two problems such as overdispersion and excess zeros will happen that can affect on variable selection and parameter estimation. Actually parameter estimation in parametric component of the semi-parametric regression model is done by penalized likelihood approach. However, in nonparametric component after local approximation using Teylor series, the estimation of nonparametric coefficients along with estimated parametric coefficients will be calculated. Using new approach leads to a properly variable selection results. In addition to representing related theories, overdispersion and excess zeros are considered in data simulation section and using EM algorithm in parameter estimation leads to increase the accuracy of end results.
Rasool Roozegar, Ali Akbar Jafari,
Volume 11, Issue 1 (9-2017)
Abstract
In this paper, we introduce a family of bivariate generalized Gompertz-power series distributions. This new class of bivariate distributions contains several models such as: bivariate generalized Gompertz -geometric, -Poisson, - binomial, -logarithmic, -negative binomial and bivariate generalized exponental-power series distributions as special cases. We express the method of construction and derive different properties of the proposed class of distributions. The method of maximum likelihood and EM algorithm are used for estimating the model parameters. Finally, we illustrate the usefulness of the new distributions by means of application to real data sets.
Reza Pourmousa, Narjes Gilani,
Volume 11, Issue 2 (3-2018)
Abstract
In this paper the mixed Poisson regression model is discussed and a Poisson Birnbaum-Saunders regression model is introduced consider the over-dispersion. The Birnbaum-Saunders distribution is the mixture of two the generalized inverse Gaussian distributions, therefore it can be considered as an extension of traditional models. Our proposed model has less dimensional parameter space than the Poisson- generalized inverse Gaussian regression model. We also show that the proposed model has a closed form for likelihood function and we obtain its moments. The EM algorithm is used to estimate the parameters and its efficiency is compared with conventional models by a simulation study. An analysis of a real data is provided for more illustration.
Mehrdad Naderi, Alireza Arabpour, Ahad Jamalizadeh,
Volume 11, Issue 2 (3-2018)
Abstract
This paper presents a new extension of Birnbaum-Saunders distribution based on skew Laplace distribution. Some properties of the new distribution are studied and the EM-type estimators of the parameters with their standard errors are obtained. Finally, we conduct a simulation study and illustrate our distribution by considering two real data example.
Afshin Fallah, Ramin Kazemi, Hasan Khosravi,
Volume 11, Issue 2 (3-2018)
Abstract
Regression analysis is done, traditionally, considering homogeneity and normality assumption for the response variable distribution. Whereas in many applications, observations indicate to a heterogeneous structure containing some sub-populations with skew-symmetric structure either due to heterogeneity, multimodality or skewness of the population or a combination of them. In this situations, one can use a mixture of skew-symmetric distributions to model the population. In this paper we considered the Bayesian approach of regression analysis under the assumption of heterogeneity of population and a skew-symmetric distribution for sub-populations, by using a mixture of skew normal distributions. We used a simulation study and a real world example to assess the proposed Bayesian methodology and to compare it with frequentist approach.
Mohammad Reza Yeganegi, Rahim Chinipardaz,
Volume 13, Issue 1 (9-2019)
Abstract
This paper is investigating the mixture autoregressive model with constant mixing weights in state space form and generalization to ARMA mixture model. Using a sequential Monte Carlo method, the forecasting, filtering and smoothing distributions are approximated and parameters f the model is estimated via the EM algorithm. The results show the dimension of parameter vector in state space representation reduces. The results of the simulation study show that the proposed filtering algorithm has a steady state close to the real values of the state vector. Moreover, according to simulation results, the mean vectors of filtering and smoothing distribution converges to state vector quickly.
Mahdi Teimouri,
Volume 14, Issue 1 (8-2020)
Abstract
The class of α-stable distributions incorporates both heavy tails and skewness and so are the most widely used class of distributions in several fields of study which incorporates both the skewness and heavy tails. Unfortunately, there is no closed-form expression for the density function of almost all of the members of this class, and so finding the maximum likelihood estimator for the parameters of this distribution is a challenging problem. In this paper, in order to tackle this issue, we propose some type of EM algorithm. The performance of the proposed EM algorithm is demonstrated via simulation and analyzing three sets of real data.
Mousa Abdi, Mohsen Madadi, Ahad Jamalizadeh,
Volume 14, Issue 2 (2-2021)
Abstract
In this article, a mixture of multivariate normal and standard exponential distributions is investigated. It is shown that the range of skewness and kurtosis coefficients for this distribution is wider than that of the skew-normal distribution. Some properties of this distribution, such as characteristic function, moment generating function, four first moments, skewness and kurtosis of distribution are presented. Also, the distribution of offine transformations and canonical forms of distribution are derived. The maximum likelihood estimation of parameters of the model is computed by using an EM algorithm. To investigate the suitability and efficiency of the model, a simulation study is presented. Finally, two numerical examples with real data sets are studied.
Farzane Hashemi,
Volume 18, Issue 2 (2-2025)
Abstract
One of the most widely used statistical topics in research fields is regression problems. In these models, the basic assumption of model errors is their normality, which, in some cases, is different due to asymmetry features or break points in the data. Piecewise regression models have been widely used in various fields, and it is essential to detect the breakpoint. The break points in piecewise regression models are necessary to know when and how the pattern of the data structure changes. One of the major problems is that there is a heavy tail in these data, which has been solved by using some distributions that generalize the normal distribution. In this paper, the piecewise regression model will be investigated based on the scale mixture of the normal distribution. Also, this model will be compared with the standard piecewise regression model derived from normal errors.
|
|
