ML2ACO: Multi-Label Feature Selection Using Multi-Layered Graph and Ant Colony Optimization

Abstract

Multi-label classification aims at finding more than one label for each instance. The performance of  multi-label classifiers is reduced when faced with high-dimensional tasks due to many irrelevant and redundant features. This work introduces a multi-layered graph modeling framework that incorporates ant colony optimization for efficient multi-label feature selection. The proposed method first maps the feature space into a multi-layer graph. In the first layer, a graph clustering method is used to correlated features. Then, a hypergraph is generated to show the relationship between layers at the second layer. We then modify the search process of the colony optimization in such a way as moving between nodes in the first layer with a high probability while searching inside the similar nodes in the second layer with a low probability. In this way, the ants selects features that are highly relevant and non-redundnant. Use a multi-label classifier in their search process  requires  high computational costs to reach the final solutions. To tackle this issue, we proposed an information-theoretic measure to assess a set of identified features. This measure computes the correlation of features with of target labels. It is also used to calculate the inner correlation between features. We evaluate the effectiveness of the proposed method on a set of real-world applications and the results show its superiority over the state-of-the-art methods.

Keywords: Feature selection, Multi-Label Data, Mutual Information, Graph Clustering, Ant Colony Optimization