Multi-label feature selection method based on dynamic weight

Multi-label feature selection attracts considerable attention from multi-label learning. Information theory-based methods intend to select the most informative features and reduce uncertain amount of information labels. Previous regard labels as constant. In fact, classification label set is captured by features, remaining uncertainty each changing dynamically. this paper, we categorize into two groups: One contains with few uncertainty, which means that respect has been obtained already-selected features; another group extensive these neglected features. Feature aims new are highly relevant in second group. Existing do not distinguish difference between groups ignore dynamic change To end, a Relevancy Ratio designed clarify under condition Afterward, Weighted defined evaluate candidate Finally, method based on (WFRFS) proposed. The experiments obtain encouraging results WFRFS comparison six thirteen real-world data sets.