On maximum depth classifiers: depth distribution approach

In this paper, we consider the notions of data depth for ordering multivariate data and propose a classification rule based on the distribution of some depth functions in . The equivalence of the proposed classification rule to optimal Bayes rule is discussed under suitable conditions. The performance of the proposed classification method is investigated in low- and high-dimensional setting using real datasets. Also, the performance of the proposed classification method is illustrated in comparison to some other depth-based classifiers using simulated data sets.