Fuzzy Federated Multi-Label Feature Selection: Reinforcement Learning and Ant Colony Optimization

Multi-label feature selection (FS) aims to reduce the dimensionality of multi-label datasets by eliminating irrelevant and redundant features, thereby improving the performance of multi-label models. However, most existing FS methods rely on centralized data, making them impractical for distributed and federated environments where resource-limited edge devices manage local datasets. Furthermore, many federated approaches assume clients have single-label data, which may not be the case in applications where instances are associated with multiple labels. To overcome these limitations, we introduce a novel federated multi-label feature selection method, leveraging fuzzy information theory combined with reinforcement learning and ant colony optimization (ACO). The method adapts fuzzy information theory to the federated setting, where clients compute fuzzy decision matrices and send them to the server, which then evaluates the associativity, interactivity, and redundancy between features. We model the multi-label feature selection process as a Markov Decision Problem and apply ACO as a multi-agent reinforcement learning approach. Features are ranked and selected based on their pheromone values. Extensive experiments on four real-world datasets across domains such as biology, images, text, and medicine show that our method surpasses existing federated and centralized multi-label feature selection techniques, achieving better results across five evaluation metrics in non-IID data distributions.