Ant colony optimization plus temporal difference and opposition-based learning for semi-supervised multi-label feature selection

This paper presents a novel feature selection method for multi-label data, combining Ant Colony Optimization with Reinforcement Learning (RL) and Opposition-Based Learning in a semi-supervised framework. Ant colony optimization navigates the feature space, while RL dynamically updates heuristic functions to predict rewards for selected features. Opposition-Based Learning enhances exploration by evaluating both original and complementary solutions. The semi-supervised learning aspect allows the algorithm to leverage the structure of the unlabeled data, improving the quality of feature selection even when labeled data is scarce. For the first time, the impact of the proposed algorithm has been systematically investigated. Compared to baseline methods, the approach achieves 7% and 3% relative improvement in Accuracy, 6% and 4% relative reduction in Hamming loss with 20% and 40% labeled data. Evaluated on 8 datasets using the Multi-Label K-Nearest Neighbors classifier, the method outperforms traditional techniques by refining feature subsets through unlabeled data. The method’s higher computational complexity and memory requirements compared to baseline techniques make it better suited for high-accuracy tasks where real-time processing is not critical, such as precision-driven biomedical analysis or batch-oriented industrial data mining. This work establishes a new benchmark for semi-supervised multi-label feature selection, prioritizing accuracy in scenarios where computational latency is tolerable.