Evaluating Conditional and Unconditional Correlations Capturing Strategies in Multi Label Classification

Abstract

In the last few years, multi label classification has attracted many scholars and researchers; due to the increasing number of modern domains that are applicable to this general type of classification. Recently, it has been believed by many researchers that the best way to handle the problem of multi label classification is by exploiting the correlations among labels. Two main strategies have been utilized to capture these correlations: conditional correlations and unconditional correlations capturing strategies. In this paper, an extensive evaluation of both strategies has been conducted, to determine the best strategy to handle multi label classification, with respect to the size of the data set and the optimized loss function. Results showed that the unconditional correlations capturing strategy overcomes the conditional correlations capturing strategy in all multi label data sets that have been used in this experiment.