An effective method of multi-label feature selection employing evolutionary algorithms

In multi-label data, each instance belongs to a set of labels, instead of one label. Due to the increasing number of modern applications for multi-label data, multi-label classification has attracted the attention of many researchers. Similar to single label data, eliminating irrelevant and/or redundant features plays an important role in improving the classifier performance. In this paper, meta-heuristic algorithms are employed to solve multi-label feature selection problem. Since the number of features in multi-label datasets is usually high, using these algorithms is not affordable in terms of computational complexity, and they may fail to find optimal feature subset. To solve this problem, irrelevant features are first removed using a filter method. Then, evolutionary algorithms are employed to find the most salient features. Experimental results demonstrate the efficiency of our proposed method compared to some existing multi-label features selection methods.