Feature Selection in Multi-label Classification based on Binary Quantum Gravitational Search Algorithm

Unlike a single-label supervisor dataset where each instance is assigned to one class label, in multi-label datasets, several class labels are assigned to each instance, which makes it difficult to build an accurate and comprehensive model from this dataset. In this study, a memetic algorithm for feature selection in a multi-label dataset is proposed. The principal innovation of this study is the offer of a novel local search algorithm which, in collaboration with binary quantum-inspired gravitational search algorithm (BQIGSA), forms the main framework of the proposed memetic algorithm. The main invention of the proposed local search algorithm is to build a number of neighbors for a solution using the prior knowledge vector and the posterior knowledge vector to select effective features and remove useless and irrelevant features. The results of implementing the proposed algorithm and comparing these results with similar works show that the proposed method in most cases leads to better results.