Abstract: Aiming at the problem that the MLKNN algorithm ignores the correlation between labels in the real world when dealing with independent labels, this paper proposes an MLKNN multi-label classification algorithm (FP-MLKNN) based on association rules. The algorithm uses association rules to mine high-order correlations between labels, and applies the association rules between labels to the MLKNN algorithm for improvement to achieve the purpose of improving the classification performance. Firstly, the MLKNN algorithm is used to obtain the characteristic confidence of the sample. Secondly, the association rule algorithm is used to mine and generate a series of strong association rules. Thirdly, the two algorithms are fused to construct a multi-label classifier to predict new labels. Experimental results show that the proposed algorithm has better classification performance than MLKNN, AdaBoostMH and BPMLL algorithms on yeast, emotions, and enron datasets, achieving a good classification effect.

Key words: multi-label classification, MLKNN, association rules；high order correlation