Multi-label classification via PCA-PLS

Compare with traditional classification, each instance in multi-label classification may be associated with multiple labels. Recent research has shown that multi-label classification is also affected by the cruse of dimensionality. Particularly, the dimensions of instance space can be much higher than those of label space. In this case, some of the existing multi-label classification methods may lose their advantages on both the efficiency and time. To overcome this problem, we propose a new method for multi-label classification. At the first step, Principal Component Analysis (PCA) is applied to reduce the dimensions of the instance space, and a new lower dimensional subspace is obtained. Next, the regression model between the new instance space and label space is structured, which is based on the Partial Least Squares (PLS), to forecast the label space of the test samples. A series of experimental results show that PCA-PLS could achieve better performance than five existing multi-label classification methods.