LAUSR

Medical diagnosis based on machine learning has received growing interest in recent years. However, traditional classification algorithms often fail to appropriately deal with medical datasets because of their high dimensionality. Manifold learning is a branch of nonlinear dimension reduction algorithms that can map the high dimensional data into a low-dimensional space. In this paper, we propose a novel manifold-based medical diagnosis algorithm named Discriminative Locally Linear Mapping (DL2M). DL2M is built on the basis of the well-known manifold leaning algorithm LLE (Locally Linear Embedding). It incorporates the discriminative information of training data into the manifold transformation of LLE, and then propagates the discriminative mapping into out-of-sample extension. DL2M is not only advantageous in preserving the local structure of original manifold, but also maps the different classes of data as far as possible in the low-dimensional feature space. The time complexity of DL2M algorithm is also discussed. Sufficient experimental results demonstrate that our method exhibits promising classification performance on the real-world medical datasets.