Recursive dependent binary relevance model for multi-label classification

Abstract

A recursive dependent binary relevance model for multilabel classification is proposed where the predicted label of a pattern is obtained in an iterative process. The motivation behind this strategy is the simultaneous intrinsic dependency of the labels and the fact that predicted labels in the final decision by themselves are estimates which can be re-estimated to improve their robustness.