ACOkEL : Ant Colony Optimization for Selecting k-Labelsets for Multi-label Classification

Multi-label classification (MLC), which allows instances to have multiple labels, has gained significant attention in recent years. The random k-labelsets ensemble (RAkEL) method improves MLC by using multiple single-label models, each created with a random subset of labels of size k. However, the random selection can miss crucial labels, affecting performance. To address this, we propose Ant Colony Optimization for selecting k-labelsets (ACOkEL). ACOkEL first clusters data using k-means, then uses an ant colony algorithm to prioritize labels based on their correlation with features within each cluster, considering the relationship between each label and other labels. This prioritization ensures the most important labels are used for classification. Experiments on nine datasets show that ACOkEL outperforms other methods by avoiding randomness and leveraging feature-label and label-label correlations, resulting in higher accuracy and better performance in multi-label classification. The proposed method demonstrates significant improvements in classification accuracy, making it a robust solution for MLC tasks.

Keywords:

Multi-Label Classification , Ant Colony Optimization , Selecting k-Labelsets , Label Powerset , Ranking Labels , Ensemble Methods