Memetic feature selection algorithm for multi-label classification

• We present a memetic feature selection algorithm for multi-label classification.
• This method employs memetic procedures to refine the feature subsets found through GAs.
• This hybridization improves the multi-label classification performance compared to counterparts.

The use of multi-label classification, i.e., assigning unseen patterns to multiple categories, has emerged in modern applications. A genetic-algorithm based multi-label feature selection method has been considered useful because it successfully improves the accuracy of multi-label classification. However, genetic algorithms are limited to identify fine-tuned feature subsets that are close to the global optimum, which results in a long runtime. In this paper, we present a memetic feature selection algorithm for multi-label classification that prevents premature convergence and improves the efficiency. The proposed method employs memetic procedures to refine the feature subsets found through a genetic search, resulting in an improvement in multi-label classification. Empirical studies using various tests show that the proposed method outperforms conventional multi-label feature selection methods.