Nonlinear Manifold Classification Based on LLE

Classification is one of the most fundamental problems in machine learning and data mining. In this paper, we propose a novel nonlinear manifold classification algorithm based on a well-known manifold leaning method called Locally Linear Embedding (LLE). LLE is a classical manifold learning algorithm, which preserves the local neighborhood structure in low-dimensional space. On the basis of LLE, our algorithm incorporates the label information of training data into the manifold mapping, so that the transformed manifold becomes more discriminative than the original manifold. The incorporated label information can help increase the similarity of homogeneous data from the same class and decrease the similarity of heterogeneous data from different classes. Sufficient experimental results demonstrate that our method exhibits better classification performance over other well-known manifold classification algorithms on seven real-world datasets.