Pattern Re-Cognitive Performance with Gaussian Mixture Model Uni-Variate and Multivariate Perspectives
Keywords:
Finite Mixture, Gaussian Mixture Model, Uni-variate Gaussian Classifier, Multivariate Gaussian Classifier, ClassificationAbstract
This study proposes supervised and unsupervised Algorithm for learning pattern identification performances with finite Gaussian mixture model from uni-variate and multivariate data sets. Bayes’ theorem for pattern discrimination was used to relate the probability density function of the data given the class to the posterior probability of the class given the data. The uni-variate case was first considered, with a continuous random variable
, which’s
, given class
, is a Gaussian with mean
and
variance. The study was extended to Multivariate analysis as the number of data dimension
. The use of finite mixture distributions to control observed and unobserved pattern heterogeneity, has become increasingly popular among those classification dynamic discrete choice models (CDDCMs). One of the challenges of using these discrete choice models is that parameters that could previously be estimated in stages must now be estimated jointly. Actual data discrimination was implemented according to Likelihood-based criterion. The intuition is that since GMM is good at describing actual complicated data, we can model the data using the GMM within one class, then use the PDF of the GMM to calculate the likelihood of any new incoming instances within every class and find the class of which the
generates the maximum likelihood. Data sets for both the uni-variate and multivariate cases are in training and test sets without missing value(s). The Gaussian discriminant model was trained by about 30% test sets of each data sets and the remaining 70% test sets was used to test the actual model pattern recognition performances. From the obtained confusion matrix, the overall proportion of test patterns correctly classified by the trained model is about 85%, attributing only 15% to classifications rate otherwise called the apparent error rate. Some vital findings and recommendations were advanced regarding the recognition accuracy of the classifiers using the training and test data sets. MATLAB Analytical software was used for all data analysis in this study.
Downloads
Published
Issue
Section
License
Copyright (c) 2024 S. I. Kalu (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.