Multi-Label Learning with Class-Based Features Using Extended Centroid-Based Classification Technique (CCBF)

Real world applications, such as news feeds categorization deal with multi-label classification problem, where the objects are associated with multiple class labels and each object is represented by a single instance (feature vector). In this paper, a new algorithm adaptation method called centroid-based multi-label classification using class-based features (CCBF) algorithm has been proposed to tackle the multi-label classification problem. It includes class-based feature vectors generation and local label correlations exploitation. In the testing stage, centroid-based classification algorithm is extended for multi-label classification problem. Experiments on reuters multi-label dataset with 103 labels demonstrate the performance and efficiency of CCBF algorithm and the result is compared with those obtained using other multi-label classification algorithms. The CCBF algorithm obtains competitive F measures with respect to the most accurate algorithms.