Abstract: A primary value oriented class label characteristic analyzing approach was proposed to extract the essential characteristics of one class label distinguishing with the others. In addition， the interpretability of label datasets could also be improved by this proposed method. Firstly， an intuitive primary value oriented class label characteristic model was built. Then， the corresponding class label characteristic extracting algorithm was designed. Finally， a classification algorithm was presented based on class label characteristic analysis. Experimental results demonstrated that the class label characteristic model can describe the characteristics of each class label for label datasets intuitively and effectively， and the given class label characteristic extracting algorithm has high execution performance. What’s more， the proposed classification algorithm has relatively high accuracy for datasets with fewer class labels.

Key words: data mining, classification, clustering, class label characteristic, primary value