Expectation Maximization Algorithm (EM): A focus on Gaussian Mixture Parameter Estimation
Keywords:
Mixture Models, Distribution of Mixture, Gaussian Mixture, Model Estimation, EM AlgorithmAbstract
Given a population of two clusters, each assumed to be normally distributed. The problem is how to estimate the parameters of these clusters including their respective mixture weights. This problem is modeled as a Gaussian mixture model, and we used the recursive iterative algorithm called Expectation Maximization (EM) to estimate these parameters. Two data sets were used in this study, namely, data Set 1and data set 2. Data Set 1 are the compiled weights of 1000 first year students of Abia State Polytechnic, Aba, Nigeria, which breaks up into 500 male students and 500 female students, while Data Set 2 are the weights of 200 babies at birth, consisting of 120 males and 80 females babies at birth from the Federal Medical Centre (FMC), Owerri, Imo State, Nigeria. We used MATLAB analytical tool to implement the EM Algorithm procedures. We also presented one of the most robust estimation/classification techniques called Maximum likelihood Estimation (MLE) via the EM algorithm for discrimination on the two component mixture of Gaussian distributions We initialized the E-Step of the Algorithm using Data Set 1&2 with the following values from table1. Dataset 1 initialization values: π1 =0.5, π2 =0.5 μ1 =75.2768, μ2 =67.9795, σ1 =6.3488, σ2 =7.8162. Dataset 2 initialization values: π1 =0.5, π2 =0.5, μ1 =3.1808, μ2 =3.2013, σ1 =0.6216, σ2 =0.5518. After full implementation of EM, we were able to optimize the initialized estimable values to: Dataset 1 optimizes values π1 =0.4994, π2 =0.5006, μ1 =75.9416, μ2 =67.3252, σ1 =6.2114, σ2 =7.2184. Dataset 2 optimized values π1 =0.4998, π2 =0.5002, μ1 =3.2953=3.0308, σ1 =0.4212, σ2 = 0.2791.
Downloads
Published
Issue
Section
License

This work is licensed under a Creative Commons Attribution 4.0 International License.
Permissible Uses and Attribution
- Citation: Any use or reproduction of content from EMSAJ Publishing Global journals must include the customary bibliographic citation, including author attribution, date, and article title.
- Permission: For commercial use or reproduction beyond fair use guidelines, explicit permission must be obtained from EMSAJ Publishing Global or the copyright holder.




