Adaptive feature selection using Label Uncertainty Reduction for multi-label classification
Gurudatta Verma, Tirath Prasad Sahu · 2024
Multi-label learning deals with datasets in which each instance is associated with a set of labels. The objective is to enhance the learning model by eliminating redundant features while assessing the uncertainty of each label. We present an approach to tackle intricate classification tasks. Our method starts by evaluating the nature of the feature space, whether binary or continuous, to ensure its suitability within the given context. Subsequently, we employ information entropy to quantify the importance of labels, which is then multiplied by the similarity between features and labels (FL) similarity. This approach effectively balances the relevance of features with the uncertainty in label prediction using Grey Relational Analysis (GRA) based optimization. Proposed method computes feature rankings by optimizing Feature Relevance and Label Certainty (FRLCO). The top-ranked features are then selected for classification tasks, and we utilize Particle Swarm Optimization (PSO) to optimize the Multi-Label k-Nearest Neighbors (MLKNN) algorithm. As a result, our proposed approach, FRLCO, excels beyond state-of-the-art multi-label feature selection techniques.