A multi-label classification using KNN and FP-growth techniques

In this paper we propose a new approach combines KNN method with FP-growth algorithm for identification and modeling existing dependencies between labels (ML-FKNN). We define and develop an algorithm that, first, utilize FP-growth algorithm for generating the association rules to identifies dependencies among the labels, then divides the whole train set into several mutually exclusive subsets to calculate the mean vectors of the each subset, and selects K nearest label neighbors for test instance by calculating its similarity with the mean vectors of the training subsets, and finally identifies the final predicted label set incorporating the discovered dependencies. Empirical evaluations on benchmark datasets shows that the proposed approach achieves high and stable accuracy results and is competitive with some existing methods for multi-label classification.