Classification performance of supervised machine learning methods on multivariate normal mixture models with application to heart failure dataset




Journal Title

Journal ISSN

Volume Title



The mixture model represents the presence of a subpopulation within a population, and it does not require the observed data set to explicitly identify the sub-populations to which each individual observation belongs. So, populations simulated from multivariate normal mixture models can be classified into sub-populations using several existing supervised machine learning models. This research generated multivariate normal mixture datasets with 1000, 2000, 5000, and 10000 observations with four groups. Each dataset had three different mean vectors, which follow the uniform distribution, and three different covariance matrices generated from the Identity, Toeplitz, and Equi-correlation matrix. Nine mean vector and covariance matrix combinations were used to generate the datasets. So, in total, 36 datasets were generated. Moreover, classification was performed using supervised machine learning methods on those datasets, and accuracy was assessed using the test dataset after portioning the datasets into training and test data. Multinomial logistic regression, support vector machine, K nearest neighborhood classifier, decision tree, bagging, and boosting algorithms were used to classify the multivariate normal mixture datasets. Heart failure is one of the major reasons of death. It brings significant health and financial burdens for patients and healthcare systems. There are several clinical and demographical risk factors of death for heart failure patients. The classification was performed on 299 heart failure patient's medical records collected at the Faisalabad Institute of Cardiology and at the Allied Hospital in Faisalabad, and the classification accuracy of the seven supervised machine learning methods was assessed.



Supervised machine learning, multivariate gaussian mixture model, simulation studies, heart failure dataset