(JPSS) Journal of Probability and Statistical Science
Use of copula for statistical classification is recent and gaining popularity. For example, statistical classification using copula has been proposed for automatic character recognition, medical diagnostic and most recently in data mining. Classical discrimination rules assume normality. But in this data age time, this assumption is often questionable. In fact features of data could be a mixture of discrete and continues random variables. In this paper, mixture copula densities are used to model class conditional distributions. Such types of densities are useful when the marginal densities of the vector of features are not normally distributed and are of a mixed kind of variables. Authors have shown that such mixture models are very useful for uncovering hidden structures in the data, and used them for clustering in data mining. Under such mixture models, maximum likelihood estimation methods are not suitable and regular expectation maximization algorithm is inefficient and may not converge. A new estimation method is proposed to estimate such densities and build the classifier based on mixture finite Gaussian densities. Simulations are used to compare the performance of the copula based classifier with classical normal distribution based models, logistic regression based model and independent model cases. The method is also applied to a real data.
Original Publication Citation
Sen, S., & Diawara, N. (2017). Supervised classification using finite mixture copula. JPSS, 15(2), 189-201.
Sen, Sumen and Diawara, Norou, "Supervised Classification Using Finite Mixture Copula" (2017). Mathematics & Statistics Faculty Publications. 205.