This paper introduces RAKEL (Random k-labELsets), an ensemble method for multilabel classification. RAKEL constructs an ensemble of Label Powerset (LP) classifiers by randomly selecting small subsets of labels and learning single-label classifiers for each subset. This approach aims to account for label correlations while avoiding the issues of LP, such as a large number of label subsets and few examples per label. The paper evaluates RAKEL on three datasets: protein function classification, semantic scene analysis, and document categorization. Experimental results show that RAKEL outperforms the Binary Relevance (BR) method and LP in terms of Hamming loss and micro-averaged F-measure. The paper also discusses the computational complexity of RAKEL and presents a unified presentation of existing evaluation measures for multilabel classification. Future work includes integrating ensemble selection methods to improve RAKEL's performance.This paper introduces RAKEL (Random k-labELsets), an ensemble method for multilabel classification. RAKEL constructs an ensemble of Label Powerset (LP) classifiers by randomly selecting small subsets of labels and learning single-label classifiers for each subset. This approach aims to account for label correlations while avoiding the issues of LP, such as a large number of label subsets and few examples per label. The paper evaluates RAKEL on three datasets: protein function classification, semantic scene analysis, and document categorization. Experimental results show that RAKEL outperforms the Binary Relevance (BR) method and LP in terms of Hamming loss and micro-averaged F-measure. The paper also discusses the computational complexity of RAKEL and presents a unified presentation of existing evaluation measures for multilabel classification. Future work includes integrating ensemble selection methods to improve RAKEL's performance.